WO2019232684A1

WO2019232684A1 - Method and device for detecting uncorrelated signal components using a linear sensor array

Info

Publication number: WO2019232684A1
Application number: PCT/CN2018/089847
Authority: WO
Inventors: Areeb RIAZ; Sebastien CURDY
Original assignee: Goertek Inc.
Priority date: 2018-06-05
Filing date: 2018-06-05
Publication date: 2019-12-12
Also published as: CN108781317B; CN108781317A

Abstract

The present invention discloses a method and device for detecting uncorrelated signal components using a linear sensor array. The method comprises: digitalizing input signals from at least three sensors in the linear sensor array within a certain time frame; buffering the digitalized signals; extracting multiple sub-bands from buffered signals corresponding to each sensor; calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors; and detecting uncorrelated signal components based on the second or more order phase difference and/or the second and/or more order magnitude ratio.

Description

METHOD AND DEVICE FOR DETECTING UNCORRELATED SIGNAL COMPONENTS USING A LINEAR SENSOR ARRAY

FIELD OF THE INVENTION

The present invention relates to the technical field of signal processing, and more specifically, to a method and device for detecting uncorrelated signal components using a linear sensor array.

BACKGROUND OF THE INVENTION

When signals from a sensor array are processed, it is better to know the sensed signals are noise or desired signals. If the sensed signals are noise, they may be discarded or labeled appropriately, and if the sensed signals are desired signals, they will be processed accordingly in relevant components of an electronic apparatus. Likewise, in array processing techniques, in order to enhance the signal of interest it is most likely required to differentiate between point sources and diffused sources. For example, in a linear microphone array, diffused noise or wind noise will impinge with different statistical characteristics along different microphones. The non-redundant information in such a situation can be used to identify whether it is the desired signal or not.

Sensors in the sensor array may be microphones, antennas and so on.

SUMMARY OF THE INVENTION

One object of this invention is to provide a new technical solution for detecting uncorrelated signal components instantaneously in the analyzed data frame using a linear sensor array.

According to a first aspect of the present invention, there is provided a method for detecting uncorrelated signal components using a linear sensor array, comprising: digitalizing input signals from at least three sensors in the linear sensor array within a certain time frame; buffering the digitalized signals; extracting multiple sub-bands from buffered signals corresponding to each sensor; calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors; and detecting uncorrelated time-frequency components based on the second or more order phase difference and/or the second and/or more order magnitude ratio.

According to a second aspect of the present invention, there is provided a device for detecting uncorrelated time-frequency components using a linear sensor array, comprising: an A/D converter, which converts input signals from at least three sensors in the linear sensor array within a certain time frame into digitalized signals; a buffer, which buffers the digitalized signals; and a processing device, which performs the following processes: extracting multiple sub-bands from the buffered signals corresponding to each sensor; calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors; and detecting uncorrelated time-frequency components based on the second or more order phase difference and/or the second and/or more order magnitude ratio.

Further features of the present invention and advantages thereof will become apparent from the following detailed description of exemplary embodiments according to the present invention with reference to the attached drawings.

BRIEF DISCRIPTION OF THE DRAWINGS

The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description thereof, serve to explain the principles of the invention.

Fig. 1 is a schematic diagram showing an estimation of the direction of arrival.

Fig. 2 shows a flow chart of a method for detecting uncorrelated signal components using a linear sensor array according to an embodiment of this disclosure.

Fig. 3 shows schematic block diagram of a device for detecting uncorrelated signal components using a linear sensor array according to an embodiment of this disclosure.

Fig. 4 shows schematic circuit diagram of the processing for detecting uncorrelated signal components using a linear sensor array according to another embodiment of this disclosure.

Fig. 5 shows schematic block diagram of the processing for detecting uncorrelated signal components using a linear sensor array according to a further embodiment of this disclosure.

Fig. 6 shows schematic block diagram of the processing for detecting uncorrelated signal components using a linear sensor array according to a still further embodiment of this disclosure.

DETAILED DESCRIPTION OF THE EMBODIMENTS

Various exemplary embodiments of the present invention will now be described in detail with reference to the drawings. It should be noted that the relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present invention unless it is specifically stated otherwise.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

Techniques, methods and apparatus as known by one of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.

In all of the examples illustrated and discussed herein, any specific values should be interpreted to be illustrative only and non-limiting. Thus, other examples of the exemplary embodiments could have different values.

Notice that similar reference numerals and letters refer to similar items in the following figures, and thus once an item is defined in one figure, it is possible that it need not be further discussed for following figures.

In an embodiment in this disclosure, uncorrelated data samples (signals) sensed by a linear sensor array in the time-frequency domain are identified. The sensor array is of a linear topology and may be placed in an electronic apparatus. The electronic apparatus may be an antenna system, a smart speaker or a cell phone, and can give a measure of the signal correlation in realistic environments. As a result, an embodiment disclosed here gives an indication of the diffused noise acoustic components, and also helps to label critical noisy scenarios such as the presence of wind and so on. This information can especially be crucial in performing speech enhancement, and/or wind noise suppression.

This disclosure takes a linear microphone array as an example to illustrate its solution. The embodiments can also be used with other linear sensor arrays in addition to the linear microphone array.

In Fig. 1, a linear array of three microphones 12-1, 12-2, 12-3 is shown. The inter-element spacing between two adjacent microphones is d, and a sound source 11 is located at a distance D, where D >> d. In this situation, the impinging waves can be considered as plane waves. In case of a wideband source, the time difference of arrival at each pair of sensors is directly related to a certain phase shift at every frequency. Consequently, there is extractable phase and/or magnitude information, which is relatable among sensor pairs. This relatable information is present both in terms of phase and magnitude. If the extracted information is redundant between pairs of sensors, the corresponding time-frequency components belong to the same point source. It is possible to use the phase and/or magnitude information already available in the signal processing chain of an electronic apparatus to determine if the received signal is desired or not. For example, in case of a Direction of Arrival (DOA) algorithm, the phase information might already be available, hence it can be re-used to determine if the signal belongs to point sources or not.

As shown in Fig. 2, at step S1100, input signals from at least three sensors in the linear sensor array within a certain time frame are digitalized.

At step S1200, the digitalized signals are buffered.

At step S1300, multiple sub-bands are extracted from buffered signals corresponding to each sensor.

At step S1400, second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors is calculated.

At step S1500, uncorrelated time-frequency components are detected based on the second or more order phase difference and/or the second and/or more order magnitude ratio.

In this embodiment, when it is determined that an input signal sensed by the sensors is wind noise or any other uncorrelated noise, a pre-defined process can be carried out to counter it. For example, if the linear sensor array is a linear microphone array, the sensed signals may be omitted without being played to a user when it is determined that the sensed signals comprise of uncorrelated components mostly. For example, while beam-forming with an antenna array or a microphone array, the White Noise Gain (WNG) in the output might be reduced by detecting uncorrelated data and consequently bypassing the signal processing chain. In addition, while using noise reduction algorithms, the information of un-correlated data can be used to build an uncorrelated data model, which subsequently can then be used to get rid of noisy uncorrelated data from the desired signals.

In an example, if uncorrelated data is detected, the signal processing of the input signals can be bypassed. For example, some components such as a FFT filter in an electronic apparatus may use input signals for training a machine learning process to get a suitable model for a point source (for example a person speaking during an enrollment process) . In such a scenario, uncorrelated noise may corrupt the desired model, leading to degradation in system performance. By appropriately labeling or discarding uncorrelated signal components, this kind of loss may be avoided.

Here, the phase difference and magnitude ratio can be used independently. Alternatively, they can be used in combination. For example, the second or more order phase difference may first be used to determine which the input signal components are uncorrelated and then the second or more order magnitude ratio may be used to determine which input signal components are uncorrelated. If anyone (or both) indicate that the input signal components match the statistical attributes of un-correlation defined by the algorithm, then uncorrelated time-frequency components are detected; otherwise, the input signal is determined as comprising of signals from desired point sources.

There are many ways to detect uncorrelated signal components. We propose detecting uncorrelated signal components using, in one example, the second order phase difference, and in second, using second order magnitude ratio. Both methods can be used independent of each other, or in combination as well.

In one example, the correlation is obtained based on the second or more order phase difference by using a first relationship table, which represents a corresponding relationship between the phase difference and the correlation, and the input signals are determined to be noise if the correlation is lower than a first predetermined threshold. For example, the first relationship table is a normalized Von-Mises distribution lookup table.

In another example, the correlation is obtained based on the second or more order magnitude ratio by using a second relationship table, which represents a corresponding relationship between the magnitude ratio and the correlation, and the input signals are determined to be noise if the correlation is lower than a second predetermined threshold. For example, the second relationship table is a normalized Exponential distribution lookup table.

The first and second predetermined thresholds may be designated by a designer. They may be obtained through experiences or through experiments, and they may be set up during manufacturing. Alternatively, they can be customized while a user is using an electronic apparatus.

For example, assuming the number of the sensors is N, N≥3, at step S1400, 2nd order phase difference and/or 2nd order magnitude ratio for the sub-bands of three sensors selected from the sensors in a linear sensor array is calculated.

Alternatively, under the above assumption, (N-1) th order phase difference and/or (N-1) th order magnitude ratio for the sub-bands of the sensors are calculated where N≥4. For example, in the case of near field (circular wave propagation) , a 3rd and 4th order may be more effective.

In the linear sensor array, the distances between two adjacent sensors from the selected three sensors are equal or in proportional relationship.

As explained above, the sensors in the linear sensor array may include microphones or antennas.

For example, at step S1300, multiple sub-bands may be extracted from each of the buffered signals, which cover a predefined bandwidth, by using a Short Time Fourier Transform technique.

Fig. 3 shows schematic block diagram of a device 50 for detecting uncorrelated signal components using a linear sensor array according to an embodiment of this disclosure. The device 50 may be used to implement the method as described above and some repetitive description will be omitted.

As shown in Fig. 3, the device 50 is connected to a linear sensor array 51 and receives sensed input signals from a linear sensor array 51. The linear sensor array 51 includes at least three sensors.

The device 50 includes an A/D converter 52, a buffer 53 and a processing device 54.

The A/D converter 52 converts input signals from at least three sensors in the linear sensor within a certain time frame into digitalized signals. The buffer 53 buffers the digitalized signals. The processing device 54 performs the following processes: extracting multiple sub-bands from the buffered signals corresponding to each sensor; calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors; and detecting noise based on the second or more order phase difference and/or the second and/or more order magnitude ratio.

For example, the processing device 54 may bypass signal processing of the input signals if mostly uncorrelated signal components are detected in the analyzed data frame.

For example, the processing device 54 further performs the following process when detecting uncorrelated signal components: obtaining the correlation based on the second or more order phase difference by using a first relationship table which represents a corresponding relationship between the phase difference and the correlation; and determining the input signals as noise if the correlation is lower than a first predetermined threshold. The first relationship table may be a normalized Von-Mises distribution lookup table, or any other statistical distribution of circular data well suited to the application.

For example, the processing device 54 may further perform the following process when detecting uncorrelated signal components: obtaining the correlation based on the second or more order magnitude ratio by using a second relationship table which represents a corresponding relationship between the magnitude ratio and the correlation; and determining the input signals are noise if the correlation is lower than a second predetermined threshold. For example, the second relationship table may be an normalized Exponential distribution lookup table.

For example, assuming the number of the sensors is N, and N=3, the processing device 54 further performs the following process: calculating 2nd order phase difference and/or 2nd order magnitude ratio for the sub-bands of three sensors selected from a linear sensor array; or calculating the (N-1) th order phase difference and/or the (N-1) th order magnitude ratio for the sub-bands of the sensors where N≥4.

In the linear sensor array, the distances between two adjacent sensors in the selected three sensors may be equal or in proportional relationship. The sensors may include microphones or antennas.

For example, the processing device 54 may further perform the following process when extracting multiple sub-bands: extracting multiple sub-bands from each of the buffered signals, which covers a predefined bandwidth, by using the Short Time Fourier Transform technique or any suitable filter bank.

Several examples will be described with reference to Figs. 4-6.

As shown in Fig. 4, the sensors 2-0-0, 2-0-1…2-0-n sense signals. The sensed signals are digitalized and buffered for a desired frame length. The signals are converted into the frequency domain, which is varying over time, using the Short Time Fourier Transform technique or any suitable filter bank. The signals have phases

where t represents time, n∈ [0, N-1] , where N is the number of sensors, ω ∈ [ω ₀, ω _K-1] spans a desired bandwidth, and K is the number of desired sub-bands. Here, the selection of the desired bandwidth may depend on the type of application, operating frequency and the frequency resolution required. For example, for a speech application, the desired bandwidth may range from 100 Hz to 4000 Hz.

As shown in Fig. 4, at the first stage, the first orders of phase differences between every two adjacent sensors are calculated at adders 2-1-0, 2-1-1…2-1-n-1. Similarly, at stage N-1, the N-1 order of phase difference

is calculated at adder 2-n.

Here, the information (phase information or magnitude information) between any adjacent pair of sensors is redundant, and the (n-1) order phase difference will be close to zero, which implies high correlation at the respective time-frequency component.

In the following, two examples of three-sensor system will be described with reference to Figs. 5, 6.

Fig. 5 shows a three-sensor system. The sensors in Fig. 5 could be antennas or microphones. In the example of Fig. 5, phase differences are used to detect uncorrelated signal components.

As shown in Fig. 5, sensors 31-1, 31-2, 31-3 sense signals and send the sensed signals to analysis modules 32-1, 32-2, 32-3. The analysis modules 32-1, 32-2, 32-3 may digitalize the sensed signals, buffer them, convert them into the time-frequency domain, extract multiple sub-bands from them and so on. At modules 33-1, 33-2, 33-3, the phases of the signals output from the analysis modules 32-1, 32-2, 32-3 are calculated as

where ω ∈ [ω ₀, ω _K-1] spans a desired bandwidth, and K is the number of desired sub-bands.

At modules 34-1, 34-2, the first orders of phase differences are calculated as

At module 35, the second order of phase difference is calculated as

Then, at stage 36, the second order of phase difference

is used as a value to fetch a phase-based correlation from a known von-Mises distribution function. The von-Mises distribution function may be implemented as a lookup table. Since in practical scenarios, the signals from the sensors may vary depending on the surrounding transfer function, the time-frequency overlap between different sources and the inherent system noise, a normalized von-Mises distribution function with a beam-width most suited for the application is preferable. However, any other distribution function for circular data may also be used.

In Fig. 5, in the graph of module 36, the horizontal axis may represent the phase difference and the vertical axis may represent the correlation per time-frequency component. Here, if the second order of phase difference is close to zero (or 360°) , the correlation is close to one. It indicates that the input signals are correlated and are likely to be the desired signals; otherwise, they may be noise.

At module 37, the detection result of uncorrelated time-frequency components or signals belonging to a point source is output. It can be used with other components of an electronic apparatus. For example, it can be used with a noise suppression/noise cancellation technique. Alternatively, if some algorithms in the signal chain are not designed to deal with uncorrelated signals, appropriate labeling can be used to bypass such algorithms.

Fig. 6 shows another three-sensor system. As explained above, the sensors in Fig. 6 could be antennas or microphones. In the example of Fig. 6, magnitude ratios are used to detect uncorrelated signal components.

As shown in Fig. 6, sensors 41-1, 41-2, 41-3 sense signals and send the sensed signals to analysis modules 42-1, 42-2, 42-3. The analysis modules 42-1, 42-2, 432-3 may digitalize the sensed signals, buffer them, convert them into the time-frequency domain, extract multiple sub-bands from them and so on. At modules 43-1, 43-2, 43-3, the magnitudes of the signals output from the analysis modules 42-1, 42-2, 42-3 are calculated as A ₁ (t, ω) , A ₂ (t, ω) , A ₃ (t, ω) , where ω ∈ [ω ₀, ω _K-1] spans a desired bandwidth, and K is the number of desired sub-bands.

At modules 44-1, 44-2, the first orders of magnitude ratios are calculated as ΔA ₂₁ (t, ω) , ΔA ₃₂ (t, ω) . At module 45, the second order of magnitude ratio is calculated as Δ ₂A ₃₁ (t, ω) .

Then, at stage 46, the second order of magnitude ratio Δ ₂A ₃₁ (t, ω) is used as a value to fetch a magnitude-ratio-based correlation from a known distribution function. The distribution function may be implemented as a lookup table, which defines the corresponding relationship between the magnitude ratio and correlation. This lookup table may be tested and tuned at the design or testing stage.

In Fig. 6, in the graph of module 46, the horizontal axis may represent the magnitude ratio and the vertical axis may represent the correlation per time-frequency component. Here, if the second order of magnitude ratio is close to one, the correlation is close to one. It indicates that the input signals are correlated and are likely to be the desired signals (representative of point sources) .

At module 47, the detection result about uncorrelated signal components and subsequently uncorrelated data frame or the desired signal components and subsequently the desired signal data frame is output. It can be used as explained above.

It will be understood by a person skilled in the prior art that a software is equivalent to a hardware except for some of the mechanical components such a speaker, a microphone and so on. In this regard, a person skilled in the art can conceive, under the teaching of this disclosure, that the processing of any of the adders 2-1-0, 2-1-1…2-1-n-1, …, 2-n in Fig. 4, the modules 32-1, 32-2, 32-3, 33-1, 33-2, 33-3, 34-1, 34-2, 35, 36, 37 in Fig. 5 and the modules 42-1, 42-2, 42-3, 43-1, 43-2, 43-3, 44-1, 44-2, 45, 46, 47 can be carried out through a hardware manner, a software manner and/or a combination thereof. For example, it can be carried out through discrete devices, ASIC, a programmable device such PLD, DSP, FPGA. Alternatively, it can be implemented in a combination of a processing device such as a CPU or a MPU and a memory, wherein instructions are stored in the memory and are used to control the processing device to performing corresponding operations. In this regard, this disclosure will not limit the implementation manners of them. A person skilled in the art can choose the implementation manners under the teaching of this disclosure in consideration of the cost, the market availability and so on.

Although some specific embodiments of the present invention have been demonstrated in detail with examples, it should be understood by a person skilled in the art that the above examples are only intended to be illustrative but not to limit the scope of the present invention.

Claims

A method for detecting uncorrelated signal components using a linear sensor array, comprising:

digitalizing input signals from at least three sensors in the linear sensor array within a certain time frame;

buffering the digitalized signals;

extracting multiple sub-bands from the buffered signals corresponding to each sensor;

calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors; and

detecting uncorrelated signal components based on the second or more order phase difference and/or the second and/or more order magnitude ratio.
The method according to claim 1, further comprising:

bypassing signal processing of the input signals if uncorrelated signal components greater than a defined threshold are detected within an analyzed time frame.
The method according to claim 1 or 2, wherein detecting uncorrelated signal components further comprises:

obtaining the correlation based on the second or more order phase difference by using a first relationship table which represents a corresponding relationship between the phase difference and the correlation; and

determining the input signals are uncorrelated if the correlation is lower than a first predetermined threshold.
The method according to claim 3, wherein the first relationship table is a normalized Von-Mises distribution lookup table.
The method according to claim 1 or 2, wherein detecting the input signals as uncorrelated further comprises:

obtaining the correlation based on the second or more order magnitude ratio by using a second relationship table which represents a corresponding relationship between the magnitude ratio and the correlation; and

determining the input signals are noise if the correlation is lower than a second predetermined threshold.
The method according to claim 5, wherein the second relationship table is a normalized Exponential distribution lookup table.
The method according to claim 1 or 2, wherein the number of the sensors is N, N≥3, and calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors further comprises:

calculating 2nd order phase difference and/or 2nd order magnitude ratio for the sub-bands of three sensors selected from the sensors in the linear sensor array; or

calculating (N-1) th order phase difference and/or (N-1) th order magnitude ratio for the sub-bands of the sensors where N≥4.
The method according to claim 7, wherein the distances between two adjacent sensors in the selected three sensors are equal or in proportional relationship.
The method according to claim 1 or 2, wherein the sensors include microphones or antennas.
The method according to claim 1 or 2, wherein extracting multiple sub-bands further comprises:

extracting multiple sub-bands from each of the buffered signals, which covers a predefined bandwidth, by using Short Time Fourier Transform technique.
A device for detecting uncorrelated signal components using a linear sensor array, comprising:

an A/D converter, which converts input signals from at least three sensors in the linear sensor array within a certain time frame into digitalized signals;

a buffer, which buffers the digitalized signals; and

a processing device, which performs the following processes:

extracting multiple sub-bands from the buffered signals corresponding to each sensor;

calculating second or more order phase difference and/or second or more order magnitude ratio for the sub-bands of the sensors; and

detecting noise based on the second or more order phase difference and/or the second and/or more order magnitude ratio.
The device according to claim 11, wherein the processing device further performs the following process:

bypassing signal processing of the input signals if uncorrelated signal components greater than a defined threshold are detected within an analyzed time frame.
The device according to claim 11 or 12, wherein the processing device further performs the following process when detecting uncorrelated signal components:

obtaining the correlation based on the second or more order phase difference by using a first relationship table which represents a corresponding relationship between the phase difference and the correlation; and

determining that the input signals are comprised of uncorrelated signal components mostly if the correlation is lower than a first predetermined threshold.
The device according to claim 13, wherein the first relationship table is a normalized Von-Mises distribution lookup table.
The device according to claim 11 or 12, wherein the processing device further performs the following process when detecting uncorrelated signal components:

obtaining the correlation based on the second or more order magnitude ratio by using a second relationship table which represents a corresponding relationship between the magnitude ratio and the correlation; and

determining that the input signals are comprised of uncorrelated time-frequency components mostly if the correlation is lower than a second predetermined threshold.
The device according to claim 15, wherein the second relationship table is a normalized Exponential distribution lookup table.
The device according to claim 11 or 12, wherein the number of the sensors is N, N≥3, and the processing device further performs the following process when calculating second or more order phase difference and/or second or more order magnitude ratio:

calculating 2nd order phase difference and/or 2nd order magnitude ratio for the sub-bands of three sensors selected from the sensors in the linear sensor array; or

calculating (N-1) th order phase difference and/or (N-1) th order magnitude ratio for the sub-bands of the sensors where N≥4.
The device according to claim 17, wherein the distances between two adjacent sensors in the selected three sensors are equal or in proportional relationship.
The device according to claim 11 or 12, wherein the sensors include microphones or antennas.
The device according to claim 11 or 12, wherein the processing device further performs the following process when extracting multiple sub-bands:

extracting multiple sub-bands from each of the buffered signals, which covers a predefined bandwidth, by using Short Time Fourier Transform technique.