Deep analysis of ILBC Packet Loss Compensation technology (PLC)

Transferred from: http://blog.csdn.net/wanggp_2007/article/details/5136609

Packet loss compensation (Packet Loss CONCEALMENT--PLC) is a very important technology in ILBC codec, and it is an indispensable part of VoIP codec applications. The ILBC PLC only handles packet compensation at the decoding end. In the decoding end according to receive the bitstream frame by the process of decoding, ILBC decoder first get each frame bitstream to determine whether the current frame is complete, if there is no problem in accordance with the normal ILBC decoding process to reconstruct the voice signal, see "deep analysis ILBC Decoder principle" If there is a loss of voice packet, then the PLC unit is processed. PLC mainly based on the decoding information of the previous frame, using the method of pitch synchronization repetition to approximate the current lost frame, in order to achieve packet loss compensation.

I. Several cases of PLC unit

1, the past frame, the current frame are receiving the correct

Entering the normal ILBC decoder decoding process, the state information of the current frame needs to be saved, including the LPC information, the decoded residual signal, etc. If the bit rate of the next frame is lost, you need to use these saved information.

2, only if the frame has dropped packets

If the current frame is not lost, then enter the PLC unit to reconstruct the LPC coefficient and residual signal. The compensation methods for LPC and residual signals are described in detail later in this article.

3, continuous multi-frame packet loss

In the case of continuous multi-frame drops, it is necessary to enter the PLC unit multiple times and take advantage of the compensated frame state information. It is important to note that the more difficult to reconstruct the lost frames in the back, the lower the gain of successive drops is reduced by frame to avoid the introduction of greater signal distortion.

4, the past frame through the PLC processing, need to be smooth with the current frame

In order to make the frame that is compensated by PLC and the frame that does not lose packet to keep the voice continuous and need to smooth, mainly according to the correlation processing of the front and back frame.

Second, the PLC reconstruction LPC coefficient

ILBC's PLC compensation for lost LPC is simply reconstructed using the LPC coefficients of the last subframe of the last frame. This method is obvious because the last sub-frame, both spatially and in time, has the most relevance to the currently lost LPC. But this simple replication will obviously introduce even greater distortion when it comes to working with multiple frames in succession.

Third, the PLC reconstruction residual signal

The excitation signal (residual signal) can usually be divided into two parts: quasi-periodic components and class-noise components. So the PLC actually needs to reconstruct these two parts first, the quasi-periodic component can be approximated according to the pitch period of the previous frame, the class noise composition can be obtained by generating random noise, and the energy proportion of the two can also draw on the proportional relation of the previous frame. So the first one is to detect the pitch of the previous frame, then the voice of the lost frame is reconstructed by the pitch synchronization, then the gain of class noise is obtained by correlation, and then the whole residual signal is reconstructed by mixing.

In the case of continuous drop frames, in order to reduce the correlation between the various compensation frames, the energy will be reduced by frame, but still produce some hearing noise. If the interpolation method, although the sound quality may be better, but will introduce a greater delay.

Iv. Disadvantages of ILBC plc

In the case of continuous drop frames, each voice frame compensated by the PLC has the same spectral characteristics (the same LPC) and the base audio rate, it is very easy to introduce a perceptible noise, especially when the pitch frequency is high, due to excessive periodicity. This problem can be mitigated by proper interpolation, but tends to introduce a greater delay.

Resources:

1, IETF:RFC3951.txt

2. The study of the advanced processing of ILBC decoding program

Deep analysis of ILBC Packet Loss Compensation technology (PLC)