Telephonic communications – Echo cancellation or suppression – Residual echo cancellation
Reexamination Certificate
2000-11-14
2004-11-09
Tieu, Binh (Department: 2643)
Telephonic communications
Echo cancellation or suppression
Residual echo cancellation
C379S406060, C379S406080, C370S289000, C370S290000
Reexamination Certificate
active
06816592
ABSTRACT:
BACKGROUND OF THE INVENTION
The invention relates to a method for echo cancellation in a digital data transmission system, in which system the end of the transmission link to which sound returns as an echo is the far end and the end of the transmission link from which an echo is reflected back is the near end, and in which a speech coding method is used on the echo path at least for a far-end signal transmitted from the far end to the near end, the method comprising the following steps: estimating the echo originating from the near end with an adaptive linear filter on the basis of the far-end signal and subtracting the echo estimate from the near-end signal transmitted from the near end to the far end.
In bi-directional data transmission networks, such as the telephone network, an echo occurs caused by the reflection of the speaker's own voice back from certain elements of the data transmission network. The echo is interfering, if there is a delay in the transmission link. A delay is usually caused by a propagation delay or digital processing of a signal.
The echo occurring in data transmission networks can be divided into two types: electric and acoustic echo. An electric echo is generated in transmission systems of transmission and reception directions of a link, such as the hybrid circuits of a telephone network (2-conductor-4-conductor converters). An acoustic echo is generated in a terminal in such a manner that the signal from the incoming transmission direction is acoustically coupled to the microphone of the transmission direction outgoing from an ear piece or loudspeaker.
In this context, the end of the transmission link to which the sound of the speaker's own voice returns as an echo is referred to as the far end, and the end of the transmission link from which the echo is reflected back is referred to as the near end.
Echo cancellers or echo suppressors are usually used to try to eliminate the echo problem. An echo canceller tries to generate an echo estimate and to cancel the echo by subtracting the echo estimate from the echo path, i.e. from the signal returning from the near end. Generally, echo estimation tries to model the impulse response of the echo path by means of an adaptive filter. In addition, non-linear processors are often used in echo cancellers to cancel the residual echo created as a result from the adaptive filtering.
An echo suppressor is usually based on comparing the power levels of a signal going out to the echo path and returning from it. If the power of the signal returning from the echo path is smaller than a certain ratio as compared to the power of the signal gone out the echo path, the transmission link returning from the echo path is disconnected so as not to let the echo through. Otherwise, the situation is interpreted as near-end speech or double speech, in which case the link can naturally not be disconnected.
Today, mainly echo cancellers are used for echo cancellation, because echo suppressors cause the following problems. Since the comparison ratio of the power levels of the far-end and near-end signals must be selected according to the worst echo situation (generally 6 dB), low level near-end speech will not get through during double speech; and although the average speech levels of the near and far end were equal, the near-end speech is occasionally cut off during double speech depending on the momentary ratio of the signal levels. Another problem is the echo during double speech. During double speech, the near-end speech gets through the echo suppressor as does the far-end echo summed to the near-end speech. The double speech echo can be reduced by attenuating the near-end and possibly also the far-end signal in the echo suppressor during double speech. However, the attenuation cannot be very strong, because it causes an interfering pumping in the speech volume.
The adaptive filters in echo suppressors use linear filters which assume that the signal returning from the echo path is both linear and time invariant (LTI, Linear Time Invariant). If this is not the case, the echo signal can be attenuated with an adaptive filter only to the extent of the linear component in the echo signal. In other words, the attenuation achieved by an adaptive filter is directly proportional to the signal-to-noise ratio of the signal returning from the echo path, i.e. inversely proportional to the non-linearity on the echo path. When the signal-to-noise ratio becomes worse, the residual echo level goes up. A non-linear processor (NLP) is often used to try to cancel this residual echo.
Data transmission networks have several sources of non-linearity. The most typical source of non-linearity in digital data transmission networks is the quantization noise generated in A/D conversion. In uniform quantization, quantization noise is, in principle, constant, whereas the signal-to-noise(distortion) ratio increases while the signal level increases. Thus, attenuation achieved on an echo signal by a linear filter is directly proportional to the momentary signal level.
In companding PCM codecs (ITU-T G.711), an analogous signal is compressed in an encoder according to a non-linear amplification curve (a or &mgr; zenith), after which the signal is uniform-quantized. Alternatively, an analogous signal can first be uniform-quantized and then non-linear-quantized according to the a or &mgr; zenith. Correspondingly, a compensating expansion of the compression is performed in a decoder. Typical of a companding PCM codec is that the signal-to-noise ratio remains almost constant on a rather wide dynamic range. In G.711 codecs, the signal-to-noise ratio is approximately 35 dB while the signal level (gaussian noise) varies between −5 dBm0 and −35 dBm0. However, on low signal levels below −35 dBm0, the signal-to-noise ratio behaves as in uniform quantization: when the signal level decreases, the signal-to-noise ratio decreases. It can thus be noted that at most an approximately 35-dB additional attenuation can be achieved on an echo signal by means of a linear filter. In practice, this attenuation is often smaller, because the level of the echo signal is rather low and thus the attenuation is dependent on the momentary signal level.
The noise summing to the echo signal can also be considered a source of non-linearity. So-called line noise is generated in analogous transmission systems. When using PCM links in digital data transmission systems, noise is not cumulated, as it is in analogous systems, and thus the main noise source is often acoustic background noise picked up by the microphone of the terminal. The attenuation of the echo signal achieved by linear filters decreases, if the line noise of the echo path or the background noise of the near end is louder than the quantization noise of the PCM codec.
A third source of non-linearity is a non-linear distortion generated in the loudspeaker of the near end, which can be considerable in a loudspeaker phone or hands-free phone. In such a case, the signal-to-distortion ratio of the returning acoustic echo has decreased as compared with the signal going out to the echo path, and the attenuation achieved by linear filters decreases correspondingly. International Patent Application PCT/US96/02073 discloses a method for compensating the non-linear distortion generated in the loudspeaker phone by modelling the distortion mechanism generated in the loudspeaker.
However, one of the most significant sources of non-linearity in digital data transmission networks is speech coding. Speech coding is today generally used in the air interface of digital mobile networks (e.g. GSM, US-TDMA, US-CDMA, PDC, TETRA). Similarly, several WLL (Wireless Local Loop) systems use speech coding in the air interface. In addition, the use of speech coding will become more common in circuit-switched PSTN networks (e.g. ITU-T, G.728, G.729, G.723.1). Speech coding will also become more common in packet switched networks (e.g. Internet calls, video conferences). It can also be noted that digital satellite mobile systems use or
Nokia Networks Oy
Pillsbury & Winthrop LLP
Tieu Binh
LandOfFree
Echo cancellation in digital data transmission system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Echo cancellation in digital data transmission system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Echo cancellation in digital data transmission system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3318409