The DTMF signal is generated by a separated source, typically a landline (corded) phone, and sent to the input lines of the module (Uplink path). The frequencies couples, sent on the voice channel , are digitised, encoded and sent by the digital transmission system. In the receiving device the signal would be reconstructed, but since the digital transmission of the voice channel is compressed and optimised for voice, this reconstruction depends on the kind of voice compression used for the transmission, and generally will not perfectly match the original signal.
There are four main types of compression for the voice channel and only the Full Rate one has no distortion, while the other three offer a different trouble level (see figures):
http://ow.ly/1gZWJ