October 09, 2005
VoIP testing
The performance of the network components is critical in deciding the success of a VoIP deployment. Applications that straddle the worlds of PSTN and IP communications utilize gateways to convert voice into IP packets and frequently these gateways need to function under high loads.
It is important that a reliable testing methodology be in place so that the quality of voice transmission can be tested. Tones can be used to test the continuity of the connections and the latency. A good VoIP testing technique should stress all the constituents of the VoIP gateway. It should be capable of conducting audio quality testing as well.
The testing methodology also needs to confirm to the quality standards followed by PSTN carriers. Stress testing is carried out test voice connections for degradation, for voice activity detection (VAD), and for complete stressing of the base signal algorithms that are used in the codecs. As VoIP is essentially a convergent technology in which voice may either originate or terminate in a PSTN network, its testing methodology should be in conformance with PSTN standards.
Testing the information signal tones for ringing, busy, etc is done for the user as well as the network equipment. In all likelihood, VoIP will become a high compression application and will have to face the same problems that wireless carriers over digital networks faced. The standards for verifying and detecting the tones were born as a result of these problems. The specifications ITU-T P.50 and P.59 relate to the use of artificial voice. In recommendation P.50, artificial voice is defined as "a signal that is mathematically defined and that reproduces the time and spectral characteristics of speech which significantly affect the performances of telecommunication systems."
The temporal behavior of human conversation, which includes pauses, mutual silence, etc, is described by recommendation P.59. This is important for the testing of the speech processing systems present in speaker phones, DCME, ATM systems, etc.
Real speech in a conversation includes pauses, noise, and variations in tone. Voice is a variable signal that also includes short portions of inter-syllabic voices. The standard speech coders such as CELP, RELP, etc use linear prediction algorithms in order to manage the variations in voice. The signals between the frequency samples are predicted by these algorithms; this allows them to manage a range of voice types with different accents.
In order to test the voice as well as the non-voice aspects of the coder architecture, a two-tier testing methodology that includes tone and artificial/real voice testing, is recommended.
--
Did you enjoy this post?
