CSR8670 - What are ANC, CVC, and DSP noise reduction?

灞波儿奔

CSR8670 - What are ANC, CVC, and DSP noise reduction? [Copy link]

1. CVC and DSP noise reduction:

When consumers are shopping for Bluetooth headsets, they always hear merchants promoting the CVC and DSP noise reduction functions of the headsets. No matter how many merchants have described this, many consumers still don't quite understand the difference between the two. In response to such a technical issue, we will today explain the working principles and differences between the two.

DSP is the abbreviation of digital signal processing. Its working principle is: the microphone collects external environmental noise, and then uses the noise reduction system inside the headset to replicate and generate a reverse sound wave equal to the external environmental noise, thereby offsetting the noise and achieving better noise reduction effect.

CVC is the abbreviation of Clear Voice Capture, which is a software noise reduction technology. Its principle is to suppress various types of reverberation noise through the built-in noise reduction software and microphone of the headset.

The difference between the two:

1. The targets are different. CVC technology mainly targets the echo generated during calls, while DSP mainly targets the high and low frequency noise in the external environment.

2. The beneficiaries are different. DSP technology mainly benefits the headset user himself, while CVC mainly benefits the other party of the call.

In conclusion, earphones that use DSP and CVC noise reduction technology can effectively reduce the noise of the external environment during calls and significantly improve the sound quality of calls and music. For example, the jetblue M8 magnetic control Bluetooth earphones newly launched by Nayin Technology have both DSP and CVC noise reduction functions, and perform very well in terms of sound quality.

2. ANC noise reduction:

ANC stands for Active Noise Control. The basic principle is that the noise reduction system generates reverse sound waves equal to the external noise to neutralize the noise. Figure 1 is a schematic diagram of a feedforward active noise reduction headset. The ANC chip is placed inside the headset. The Ref mic (reference microphone) is on the headset earmuff and collects ambient noise. The Error mic (Error Microphone) is inside the headset and collects the residual noise after the noise reduction process. The speaker plays the anti-noise after ANC processing.

Figure 1

download attach  save to album

2019-10-12 16:24 上传

Figure 2

　　Figure 2 is a schematic diagram of the ANC system, which has three layers separated by dotted lines. The top layer, the primary path, is the acoustic path from the ref mic to the error mic, and the response function is represented by ; the middle layer is the analog channel, in which the secondary path is the path from the output of the adaptive filter to the returned residual, including DAC, reconstruction filter, power amplifier, speaker playback, re-acquisition, pre-amplifier, anti-aliasing filter, ADC; the bottom layer is the digital path, in which the adaptive filter continuously adjusts the filter weight coefficients to reduce the residual until convergence. The most commonly used solution is to use an FIR filter combined with an LMS algorithm to implement an adaptive filter. Simplifying Figure 2 gives Figure 3.

Figure 3

　　Let's first briefly talk about the principles of adaptive filters and LMS (Least mean square) algorithms, and then talk about Figure 3. As shown in Figure 4, given an input and a desired output, the adaptive filter will update the coefficients each iteration so that the difference between its output and the desired output becomes smaller and smaller, until the residual is close enough to 0 and converges. LMS is an update algorithm for adaptive filters. The objective function of LMS is the square of the instantaneous error. In order to minimize the objective function, gradient descent is applied to it to obtain the update formula of the algorithm. (This algorithm idea of using gradient descent to minimize an objective and thus obtain the update formula of the desired parameter is very common, such as linear regression.) The update formula of the LMS algorithm using FIR filters is: , where is the step size. If the size is adjusted with iteration, it is a variable step size LMS algorithm.

Figure 4

　　Let's talk about Figure 3. Here, the output of the adaptive filter has to pass through before it is compared with the desired output, which will cause instability. In the words of the literature, "the error signal is not correctly 'aligned' in time with the reference signal", which destroys the convergence of LMS. (I still don't understand what this means T__T) An effective method is FXLMS (Filtered-X LMS), which is to let x(n) pass through before inputting it into the LMS module, which is an estimate. The objective of FXLMS is:

So gradient = , where is unknown and is approximated by its estimate, so the update formula of FXLMS is where .

　　When the adaptive filter converges, , therefore. In other words, the weight coefficients of the adaptive filter are determined by the primary path and secondary path of the headphones. The primary path and secondary path of the headphones are relatively stable, so the weight coefficients of the adaptive filter are also relatively stable. Therefore, in order to simplify the implementation, the weight coefficients of the ANC headphones of some manufacturers are determined at the factory. Of course, the listening experience of this kind of ANC headphones is obviously not as good as that of ANC headphones with true adaptive meaning, because in actual situations, factors such as the direction of external noise relative to the headphones and different temperatures will affect the channel response of the headphones.