3.6.1Gateway No.1
“Evaluation Module”
|
|
Parameter
|
Reference
|
Gateway
|
Echo Loss (G.122)
|
48.4 dB
|
48.3 dB
|
The echo attenuation under single talk conditions is high. The echo loss result is only limited by the idle noise. Furthermore the echo attenuation is constant vs. time (fig. 5.43), the Relative Approach does not detect any echo components in figure 5.44.
|
|
|
The near end bursts are completely transmitted even under worst case conditions (fig. 5.45). The speech double talk sequence is also completely transmitted, gaps do not occur in fig. 5.46. The echo attenuation is constant over the whole double talk analysis (5.47).
|
Fig. 5.43: Echo att., 5 dBm0
|
Fig. 5.44: Echo RA., 16 dBm0
|
|
|
|
Fig. 5.45: Simulated DT (enlar.)
|
Fig. 5.46: Near end speech
|
Fig. 5.47: Echo att., DT
|
The 4-wire detection works reliable. The near end background noise is not attenuated or disturbed by implemented NLP or echo suppression if a far end signal is applied (fig. 5.48). The homogeneous color in the Relative Approach analysis indicates an undisturbed background noise transmission. Basically the same can be derived from the 2D RA analysis in fig. 5.50. Echo components are not detected, background noise is completely transmitted.
|
|
|
|
|
Fig. 5.48: Noise transmission with far end signal
|
Fig. 5.49: Δ Relative Approach Café
|
Fig. 5.50: Δ RA, Noise transm. with far end speech
|
| 3.6.2Gateway No.2
“Complete System”
|
|
Parameter
|
Reference
|
Gateway
|
Echo Loss (G.122)
|
48.4 dB
|
59.0 dB
|
The echo loss result of 59 dB under single talk conditions is very high and indicates a completely muted send signal (digital zero). This is also detected by the Relative Approach in fig. 5.52. The analysis leads to a “zero” signal during the presence of the two sentences in this analysis.
|
|
|
The near end double talk bursts are completely transmitted in fig. 5.53, short gaps occur in the transmitted speech signal under double talk conditions in fig. 5.54. The high echo attenuation at the beginning of the analysis in fig. 5.55 indicates that NLP is partly active.
|
Fig. 5.51: Echo att., 5 dBm0
|
Fig. 5.52: Echo RA., 16 dBm0
|
|
|
|
Fig. 5.53: Simulated DT (enlar.)
|
Fig. 5.54: Near end speech
|
Fig. 5.55: Echo att., DT
|
The 4-wire detection is not active under this test condition. Near end background noise is completely suppressed by NLP coincident to the application of a far end signal (fig. 5.56). The dark color in the Relative Approach analysis shows that the signal characteristic is completely lost. The strong variation in the 2D RA analysis in fig. 5.58 demonstrates the unexpected modulation, if the signal is completely suppressed (negative scaling) and partly transmitted. The transition from silence periods to background noise transmission leads to sudden positive amplitudes in fig. 5.58. This is unexpected for the human ear.
|
|
|
|
Fig. 5.56: Noise transmission with far end signal
|
Fig. 5.57: Δ Relative Approach Café
|
Fig. 5.58: Δ RA, Noise transm. with far end speech
| 3.6.3Gateway No.3
“Evaluation Module”
|
|
Parameter
|
Reference
|
Gateway
|
Echo Loss (G.122)
|
48.4 dB
|
48.3 dB
|
The echo loss result is comparable to the references connection with gateways and only limited by the idle noise in the test setup. The echo attenuation is constant vs. time (fig. 5.59). The idle noise is not modulated, the Relative Approach in fig. 5.60 does not detect any unexpected artifact for the human ear.
|
|
|
All near end double talk bursts are completely transmitted (see fig. 5.61). The same can be analyzed for the speech double talk sequence in fig. 5.62. The echo attenuation is high and constant over the whole double talk sequence as analyzed in fig. 5.63
|
Fig. 5.59: Echo att., 5 dBm0
|
Fig. 5.60: Echo RA., 16 dBm0
|
|
|
|
Fig. 5.61: Simulated DT (enlar.)
|
Fig. 5.62: Near end speech
|
Fig. 5.63: Echo att., DT
|
The non-linear processor or echo suppression unit attenuates the send signal coincident to the application of a far end signal in fig. 5.64. The resulting gaps are masked by comfort noise. Although the signal characteristic changes (café noise vs. comfort noise) the Relative Approach analysis shows that the transmission characteristic is still relatively smooth. Comfort noise adapts on the background noise signal level. The 2D RA analysis in fig. 5.66 also detects the modulation caused by the transition from original background noise to comfort noise but the analysis amplitude is still moderate.
|
|
|
|
|
Fig. 5.64: Noise transmission with far end signal
|
Fig. 5.65: Δ Relative Approach Café
|
Fig. 5.66: Δ RA, Noise transm. with far end speech
|
| 3.6.4 Gateway No.4
“Complete System”
|
|
|
Parameter
|
Reference
|
Gateway
|
|
Echo Loss (G.122)
|
48.4 dB
|
48.5 dB
|
|
The echo is completely suppressed under single talk conditions. The measured result is only limited by the idle noise level. The echo attenuation is constant vs. time (fig. 5.67) and the hearing model based Relative Approach does not detect any disturbing components in the sending signal (fig. 5.68).
|
|
|
|
The near end bursts are completely transmitted even under worst case conditions (fig. 5.45). The speech double talk sequence is also completely transmitted, gaps do not occur in fig. 5.70. The echo attenuation is constant over the whole double talk analysis (5.71).
|
|
Fig. 5.67: Echo att., 5 dBm0
|
Fig. 5.68: Echo RA., 16 dBm0
|
|
|
|
|
|
|
Fig. 5.69: Simulated DT (enlar.)
|
Fig. 5.70: Near end speech
|
Fig. 5.71: Echo att., DT
|
|
The analysis in fig. 5.72 indicates a slight level mismatch between the undisturbed transmission and the transmission coincident to the application of a far end signal. The near end background noise is attenuated by implemented NLP but adaptive comfort noise is inserted to mask the gaps. Comfort noise level matches quite good, the disturbances are limited as indicated by the 3D Relative approach in fig. 5.73. The 2D RA analysis in fig. 5.74 shows that the noise transmission is transparent during the second sentence.
|
|
|
|
|
Fig. 5.72: Noise transmission with far end signal
|
Fig. 5.73: Δ Relative Approach Café
|
Fig. 5.74: Δ RA, Noise transm. with far end speech
| 3.6.5 Gateway No.5
“Complete System”
|
|
|
Parameter
|
Reference
|
Gateway
|
|
Echo Loss (G.122)
|
48.4 dB
|
48.8 dB
|
|
The echo attenuation expressed by the one dimensional echo loss result is sufficiently high. However, further analyses like the echo attenuation vs. time (fig. 5.75) demonstrate that the residual echo signal fluctuates vs. time.
|
|
|
|
The near end bursts are completely transmitted during double talk (fig. 5.77) but echo components appear during the pauses between two bursts. The speech double talk sequence is also impaired by short gaps (fig. 5.78). The echo attenuation is not constant during double talk (fig. 5.79).
|
|
Fig. 5.75: Echo att., 5 dBm0
|
Fig. 5.76: Echo RA., 16 dBm0
|
|
|
|
|
|
Fig. 5.77: Simulated DT (enlar.)
|
Fig. 5.78: Near end speech
|
Fig. 5.79: Echo att., DT
|
|
The implemented echo suppression interrupts the near end background noise if a far end signal is applied (fig. 5.80). The dark color in the Relative Approach analysis shows that the signal characteristic is completely lost during the presence of a CS signal burst. The strong variation in the 2D RA analysis in fig. 5.82 also demonstrates the unexpected modulation. Negative values indicate missing signal features. The transition from silence periods to background noise transmission leads to positive amplitudes in fig. 5.58. This is unexpected for the human ear.
|
|
|
|
|
Fig. 5.80: Noise transmission with far end signal
|
Fig. 5.81: Δ Relative Approach Café
|
Fig. 5.82: Δ RA, Noise transm. with far end speech
| 3.6.6Gateway No.6
“Complete System”
|
|
Parameter
|
Reference
|
Gateway
|
Echo Loss (G.122)
|
48.4 dB
|
52.6 dB
|
The echo loss result of 53 dB is higher than for the reference and indicates a muted send signal under this condition (digital zero). The Relative Approach in fig. 5.84 detects idle noise during the application of the two test signals. However, the level is low enough.
|
|
|
Strong echo components appear during the pauses between two near end signal bursts in fig. 5.85. The speech double talk sequence is also impaired by short gaps (fig. 5.86). The echo attenuation strongly varies during the double talk sequence (fig. 5.87).
|
Fig. 5.83: Echo att., 5 dBm0
|
Fig. 5.84: Echo RA., 16 dBm0
|
|
|
|
Fig. 5.85: Simulated DT (enlar.)
|
Fig. 5.86: Near end speech
|
Fig. 5.87: Echo att., DT
|
Background noise present at the near end is completely suppressed by NLP coincident to the application of a far end signal (fig. 5.88). The dark color in the Relative Approach analysis shows that the signal characteristic is completely lost if the CS signal is active in receiving direction (fig. 5.89). The strong variation in the 2D RA analysis in fig. 5.90 demonstrates the unexpected modulation for the human ear.
|
|
|
|
|
Fig. 5.88: Noise transmission with far end signal
|
Fig. 5.89: Δ Relative Approach Café
|
Fig. 5.90: Δ RA, Noise transm. with far end speech
|
| 3.6.7Gateway No.7
“Complete System”
|
|
|
Parameter
|
Reference
|
Gateway
|
|
Echo Loss (G.122)
|
48.4 dB
|
53.0 dB
|
|
The echo loss result of 53 dB measured for this gateway implementation is also higher than for the reference connection. The echo attenuation is stable vs. time. The Relative Approach in fig. 5.84 detects a slightly different signal characteristic if the speech sequences are applied of not (fig. 5.92).
|
|
|
|
Echo components occur in fig. 5.93 during the pauses between two near end signal bursts. Gaps in the double talk speech signal are caused by NLP (fig. 5.94). The NLP is very quickly inserted at the beginning and at the end of the double talk sequence (fig. 5.95).
|
|
Fig. 5.91: Echo att., 5 dBm0
|
Fig. 5.92: Echo RA., 16 dBm0
|
|
|
|
|
|
|
Fig. 5.93: Simulated DT (enlar.)
|
Fig. 5.94: Near end speech
|
Fig. 5.95: Echo att., DT
|
|
The non-linear processor or echo suppression unit attenuates the send signal coincident to the application of a far end signal in fig. 5.64. The time constants relevant for the loss insertion are short, the signal characteristics varies too quickly. The Relative Approach analysis in fig. 5.97 shows the resulting disturbances. The 2D RA analysis in fig. 5.98 also detects these modulations caused by the quickly transition from original background noise transmission to comfort noise insertion.
|
|
|
|
|
Fig. 5.96: Noise transmission with far end signal
|
Fig. 5.97: Δ Relative Approach Café
|
Fig. 5.98: Δ RA, Noise transm. with far end speech
|
Share with your friends: |
