D.1 Test procedure

26.1323GPPRelease 18Speech and video telephony terminal acoustic test specificationTS

As speech test signal, the second sentence of the first female speaker (female1.wav) of the English test sentences according to ITU-T P.501 is used. . When measuring in receiving direction the signal is pre-filtered according to the used bandwith and normalized to an active speech level of -16dBm0. When measuring in sending direction the signal is calibrated to an active speech level of -4.7 dBPa at MRP. For electrical interface UE, the active speech level is calibrated to -60 dBV for analogue and to -16 dBm0 for digital connections.. The sequence is centered within a 4.0s time window and repeated 40 times, resulting in a test file of 160.0s length. Alternatively CSS signals may be used with the corresponding adaptation of the step size to 333ms. The test signal used shall be reported.

NOTE– Care should be taken to apply the test procedure to the receive direction, since the measurement could lead to biased results due to jitter buffer adaptation of the DUT during the test.

The clock skew is determined as follows:

– The variable delay per time step is calculated as the time shift of the maximum of the cross correlation between measured and source signal for each time step.

– The step size of the calculation of the variable delay is 4.0s for the above test signal, which corresponds to one single sentence of the source signal.

– This analysis provides the output signal Y(tn) consisting of one single delay value for each time step, i.e., for each sentence of the above source signal.

– A median filter of length 3 is applied to the delay vs. time series Y(tn) yielding the time series Ŷ(tn). This smoothes small spikes without soften the delay jumps.

– Delay jumps due to jitter buffer over- or underrun are detected as the time steps where the absolute delay difference abs(Ŷ(tn+1) – Ŷ(tn)) is larger than a threshold ΔYmax, which is appropriate to detect a maximum clock skew of 500 PPM.

– A "continuous" delay vs. time series is generated where the delay jumps are eliminated, e.g., by cutting out the delay jumps and "stitching" together the jump-free segments.

– The clock skew is estimated as the regression coefficient β of the "continuous" delay vs. time series in parts per million (PPM):

In case of clock skews higher than 50 PPM the accuracy of the cross-correlation calculation decreases. In such event it is recommended to take a two-step approach. In the first step a coarse skew compensation is performed and the clock of the reference client is adjusted to the PPM derived in the first step. A fine compensation is performed in the second step by applying the same procedure. The final clock skew in PPM is the sum of both clock skews.