SWITCHED-GEOMETRY MICROPHONE ARRAY ARRANGEMENT AND METHOD FOR PROCESSING OUTPUTS FROM A PLURALITY OF MICROPHONES
Field of the Invention
This invention relates to microphones for sound (e.g., speech) reception.
Background of the Invention
In the field of this invention it is known that arrays of microphones, whose output signals are processed and then combined into a single output signal, can enhance sound received from one particular source location (or direction) , relative to sounds from other locations (or directions) . An important application is speech reception in noisy and/or reverberant environments, when it is not feasible to position a microphone very close to the talker's mouth. For example, hands-free telephones in cars and palmtop computers, held at arm's length, can receive speech more clearly by using arrays rather than single microphones.
The spatial response pattern of an array, which makes it sensitive to sound from the desired source location but insensitive to sound coming from other locations, depends on the spatial layout of its microphones. Hence, when the locations of the sound sources change, the microphone layout would ideally be adjusted too, so as to maintain the optimal spatial response. Significant motion of the speech and noise sound sources relative to the array's platform is likely in many applications (e.g., hand-held devices, when the user moves his/her head or hand) .
However, moveable microphones, which automatically changed their layout within their platform, would clearly be too costly and bulky for most applications.
Another way of compensating for motion of the sound sources relative to the array platform is by adjusting the array's signal processing, and this is quite adequate for some types of array. For example, the spatial response pattern of a delay-and-sum array can be made to track a moving source by adjusting the delays on the microphones' outputs. In other types of array, however, the spatial response is highly constrained by the microphone layout. For example, to obtain a highly directional response with a compact array, the microphones must be placed along a line roughly parallel to the desired source direction.
A compact array is one whose inter-microphone distances are much smaller than the sound wavelengths they receive. This is the only kind of array that will fit into many small devices, such as hand-held devices. Hence, a compact array whose microphones moved in response to relative motion of the sound sources would have many potential applications.
An array's microphones may effectively be moved by switching its input channels between microphones fixed at different locations. Such switching of microphone output signals may be achieved in various known ways. Normally, microphone arrays employ digital signal processing, as this is much more flexible than analogue signal processing, and allows better adaptation to
changing acoustic conditions. In such a "fully digitally processed" microphone array, the analogue output signal from each microphone is passed through an analogue-to- digital converter (ADC) and then a digital signal processor (DSP) , and the processed signals are then summed to give the array's output. A performance evaluator continually measures the array's output signal quality (e.g., its Signal/Noise Ratio - S R) , and this performance measure is then fed back to an adaptation processor, which adjusts the digital signal processors so as to optimise performance. Thus, the array adapts to changing acoustic conditions. The microphone layout in such a "fully digitally processed" array could effectively be varied by varying the overall gains of the DSP channels (for example, a two-microphone array would be obtained when the adaptation processor set the gains in all but two channels to zero) . By changing the pair of channels that remains active, the array's alignment could effectively be rotated. Arrays of this kind are known from U.S. Patent 5,715,319. Appended Figure 1 illustrates a typical example.
Clearly, the more microphones there are in an array, the more different layouts may be created by selecting subsets of the microphones. This can allow the array layout to be brought closer to the ideal layout for any given acoustic conditions. However, the cost, the physical dimensions and the power consumption of the array also rise with the number of microphones, and the number of ADC's and DSP's associated with them. The digital processing circuitry accounts for a large
fraction of the cost and power consumption of a microphone array.
In a compact array, whose microphones are clustered within a small volume, the digital processing circuitry also accounts for a large part of the array's physical size. This is especially true when the microphones are micro electromechanical (MEMS) devices, all integrated onto one silicon chip. Using such MEMS technology, it is possible to create a compact array that can be adopted into a very wide range of layouts, because it has a very large number of very small microphones.
It is thus desirable to reduce the amount of digital processing circuitry used by an array, without reducing its number of microphones, which would limit the range of microphone layouts it could adopt. One way to reduce the cost, size and power of the digital circuitry is to replace discrete components with a reduced number of integrated circuits.
Another known way of decreasing the cost and size of the processing circuitry is to use multiplexing (i.e., processing multiple channels one after another in the same circuit) . Clearly, however, multiplexing requires faster digital circuitry, and there is a limit to the number of microphones that can be multiplexed into a single ADC or DSP. Appended Figure 2 illustrates a typical example.
An alternative to digitising and then processing every microphone output signal is available. The first part of
the alternative technique is to initially process the microphone outputs in analogue circuitry, so as to generate a reduced number of analogue signals. The second part of the alternative technique digitises and further processes the reduced number of analogue signals. This approach requires less and/or slower digital circuitry. Additional analogue pre-processing circuitry is required, but provided this is relatively simple, it can have lower cost, size and power consumption than the digital circuits it replaces. Appended Figure 3 illustrates a typical example .
In one type of analogue pre-processed array, each microphone output is passed through a fixed analogue filter, certain sets of the filtered signals are then summed, and the resultant reduced number of signals is then digitised and processed further. In an example of this approach also known from U.S. Patent 5,715,319, the filters on a summed set of microphones each have a different pass band, effectively creating a virtual microphone whose effective location varies with sound frequency. However, this type of analogue pre-processed array has a fixed set of microphones connected to each of its DSP channels, which limits the range of microphone layouts it can achieve.
A need therefore exists for a switched-geometry microphone array and method wherein the abovementioned disadvantage (s) may be alleviated.
US-A-5848170 and DE-A-3007585 describe prior art arrangements .
Statement of Invention
In accordance with a first aspect of the present invention there is provided a switched-geometry arrangement, for processing outputs from a plurality of microphones, as claimed in claim 1.
In accordance with a second aspect of the present invention there is provided a switched-geometry microphone arrangement as claimed in claim 7.
In accordance with a third aspect of the present invention there is provided a method, for processing outputs from a plurality of microphones, as claimed in claim 9.
Brief Description of the Drawing (s)
Two switched-geometry microphone array arrangements incorporating the present invention will now be described, by way of example only, with reference to the accompanying drawing (s), in which:
FIG. 1 is a block-schematic diagram illustrating a known "fully digitally processed" microphone array arrangement;
FIG. 2 is a block-schematic diagram illustrating a known multiplexed digitally processed microphone array arrangement ;
FIG. 3 is a block-schematic diagram illustrating a known analogue pre-processed microphone array arrangement;
FIG. 4 is a block-schematic diagram illustrating a generic switched-geometry microphone array arrangement incorporating the invention; and
FIG. 5 is a block-schematic diagram illustrating a preferred embodiment of a switched-geometry microphone array arrangement, based on that shown in FIG. 1, incorporating the invention.
Description of Preferred Embodiments
A "fully digitally processed" microphone array employs digital signal processing, as this is much more flexible than analogue signal processing, and allows better adaptation to changing acoustic conditions. As shown in FIG. 1, in a conventional microphone array of this type the analogue output signal from each microphone (10.1- 10.7) is passed through an analogue-to-digital converter (13.1-13.7) and then a digital signal processor (14.1-
14.7) . The processed signals are then summed at an adder (15) to give the array's output 0. A performance evaluator (16) continually measures the array's output signal quality (e.g. its Signal/Noise Ratio - SNR) . This performance measure is then fed back to an adaptation processor (17) , which adjusts the digital signal processors (14.1-14.7) so as to optimise performance. By this means, the array adapts to changing acoustic conditions .
The microphone layout in such a "fully digitally processed" array could effectively be varied by varying
the overall gains of the DSP channels. For example, a two-microphone array would be obtained when the adaptation processor set the gains in all but two channels to zero, as illustrated in FIG. 1. By changing the pair of channels that remains active, the array's alignment (19) could effectively be rotated. It will be appreciated that, as illustrated, the arrow 19 shows array alignment with only DSP's 14.3 and 14.7 active.
Clearly, the more microphones there are in an array, the more different layouts may be created by selecting subsets of the microphones. This can allow the array layout to be brought closer to the ideal layout for any given acoustic conditions. However, the cost, the physical dimensions and the power consumption of the array also rise with the number of microphones, and the number of analogue-to-digital converters (ADC's) and digital signal processors (DSP's) associated with them. The digital processing circuitry typically accounts for a large fraction of the cost and power consumption of a microphone array.
In a compact array, whose microphones are clustered within a small volume, the digital processing circuitry typically also accounts for a large part of the array's physical size. This is especially true when the microphones are micro electromechanical (MEMS) devices, all integrated onto one silicon chip. Using such MEMS technology, it is possible to create a compact array that can adopt a very wide range of layouts, because it has a very large number of very small microphones.
It is thus desirable to reduce the amount of digital processing circuitry used by an array, without reducing its number of microphones, which would limit the range of microphone layouts it could adopt. One way to reduce the cost, size and power of the digital circuitry is by replacing discrete components with a reduced number of integrated circuits. Another known way of decreasing the cost and size of the processing circuitry is to use multiplexing, i.e., processing multiple channels one after another in the same circuit .
Referring now also to FIG. 2, in a known multiplexed microphone array, during each sample period the multiplexer (21) connects the output of each microphone (20.1-20.7) in turn to a single ADC (23). The ADC and the DSP (24) process the latest signal samples from each microphone one after another, before combining them to generate the latest array output signal sample. Clearly, multiplexing requires faster digital circuitry, and there is a limit to the number of microphones that can be multiplexed into a single ADC or DSP.
An alternative to digitising and then processing every microphone output signal is to initially process the microphone outputs in analogue circuitry, so as to generate a reduced number of analogue signals, which are then digitised and processed further. This approach requires less and/or slower digital circuitry. Additional analogue pre-processing circuitry is required, but provided this is relatively simple, it can have lower cost, size and power consumption than the digital circuits it replaces.
Referring now to FIG. 3, in one known type of analogue pre-processed array, each microphone output (30.1-30.7) is passed through a fixed analogue filter (31.1-31.7), and certain sets of the filtered signals are then summed at adders (32.1-32.3) . The resultant reduced number of signals is then digitised by ADC's (33.1-33.3) and processed further. In a published example of this approach (also known from US Patent 5,715,319) the filters on a summed set of microphones each have a different pass band, the effect of which is to create a virtual microphone whose effective location varies with sound frequency. However, this type of analogue pre- processed array has a fixed set of microphones connected to each of its DSP channels, which limits the range of microphone layouts it can achieve.
Referring now to FIG. 4, in a general form of the present invention each of M microphones (40.1-40.M) in an array passes its output signal to a respective analogue switch (41.1-41.M). Each switch passes its input signal to one of N outputs (N ≤ M) , or does not pass it on (it will be appreciated that the lowest output line from each of the switches represents no output from that switch, i.e., switching off the switch' s respective microphone) . The nth output signals (for n = 1 to N) of all the switches are summed, via a chain of analogue adders (42.2. - 42.M.n), to give the nth input to an array signal processor (43) . The array signal processor processes and combines its N input signals so as to generate the array's final output O.
By adjusting the analogue switches (41.1-41.M) , the combination of microphone outputs that is passed to each input of the array signal processor (43) may be varied. Just one microphone at a time could be connected to each signal-processor input, leaving M-N microphones unused. Alternatively, more than one microphone could be connected to some or all of the signal processor inputs . The signal on each signal processor input is the sum of the output signals of all the microphones connected to it.
It will be understood that the combination of its system of analogue switches (41.1-41. M) and adders (42.2.1- 42.M.N), particularly the way in which the layout of microphones connected to each array signal processor input can be varied, allows the microphone array (40.1- 40.M) in the arrangement of FIG. 4 to attain a much wider range of effective microphone layouts, permitting adaptation of its spatial response pattern to changing acoustic conditions.
One particular embodiment of this invention is illustrated in FIG. 5. In this embodiment, an analogue pre-processor (51) includes switches and adders, and an array signal processor (53) comprises three ADC's (53.1- 53.3), each feeding a respective DSP (54.1-54.3), whose outputs are summed at an adder (55) to give the array's final output O. A performance evaluator (56) continually measures the array's output signal quality. This performance measure is then fed back to an adaptation processor (57) , which adjusts the digital processors (54.1-54.3) and the switches within the analogue preprocessor (51) so as to continually optimise performance.
In this embodiment there are seven microphones (50.1- 50.7), arranged at the corners and centre of a hexagon as shown in FIG. 5. For the purpose of this example, the switches within the analogue pre-processor (51) divide these microphones into three sets (50.1 & 50.3; 50.2, 50.4 & 50.6; and 50.5 & 50.7) and the adders within the analogue pre-processor (51) sum the outputs of each set. The three summed signals are then fed to the ADC's (53.1- 53.3) . The effect of summing a set of microphone outputs is to create a single virtual microphone, located at the centroid of the real microphones. This embodiment thus creates an array of three virtual microphones, aligned as shown by the arrow (59) . It will be understood that, as illustrated, the arrow 59 shows array alignment with microphones switched to array processor inputs as follows: microphones 50.1 and 50.3 switched to input 1; microphones 50.2, 50.4 and 50.6 switched to input 2; and microphones 50.5 and 50.7 switched to input 3. Clearly, the effective layout of the array may be varied by changing the sets of microphones that are summed together on each ADC input .
It will be understood that the advantage of adaptive control is that the array's output signal quality is kept high despite changing acoustic conditions. For example, the array's spatial response pattern may be adjusted to track moving sound sources .
When the adders within the analogue pre-processor (51) sum the outputs of microphones that are close together, this has two advantages.
Firstly, adding microphone outputs creates a virtual microphone at the microphones' centroid. Virtual microphones may be created at varying locations by switching different sets of microphone outputs to an adder. This gives greater flexibility to vary the effective configuration of microphones connected to the array processor, and hence optimise array performance.
Secondly, adding microphone outputs reduces microphone intrinsic noise, relative to sound signals. Provided the microphones are close together, their sound signals are very similar, and add coherently, whereas their intrinsic noise signals are incoherent. Microphone intrinsic noise is a significant problem in very small microphones, such as MEMS microphones, so adding the outputs of many adjacent MEMS microphones can improve array performance.
Also, it will be appreciated that fixed analogue filters may additionally be included. Such fixed analogue filters filter microphone output signals before they are fed to the analogue adders. When filters with different frequency responses are applied to the inputs to an adder, the location and sensitivity of the resultant virtual microphone varies with frequency. This allows the effective configuration of microphones connected to the array processor to vary with frequency, which gives greater flexibility to optimise array performance over a range of frequencies.
Further, it will be appreciated that generally the array processor (43) may use analogue or digital signal
processing, which may be fixed or adaptive, as appropriate to the application.
It will be understood that the switched-geometry microphone array described above provides the following advantages :
In theory, a microphone array with the same performance as any embodiment of this invention could be realised by passing each microphone output straight to an ADC, and performing all additions and other signal processing in digital circuitry. In many cases, however, the simple analogue switches and adders used in the embodiments of the invention described above will have lower cost, size and power consumption than their digital alternative.
This invention may be most advantageous when a large number of microphones (-fifty or more) is connected to a much smaller number of array signal processor inputs, and especially when the analogue switches and adders are integrated into a silicon chip along with the microphones .
Since each microphone has a similar set of switches and adders on its output, this set could be placed next to each microphone, to form a "microphone block". A regular array of any size could then be built up by connecting identical blocks together. This block-based structure is particularly suited to arrays integrated into a single silicon chip.
This invention facilitates summing the outputs of a large number of adjacent microphones to provide each input to an array signal processor, rather than having just one microphone on each input . The main advantage of summing the outputs of a large number of adjacent microphones is that microphone intrinsic noise is reduced, relative to sound signals. Since the microphones are close together, their sound signals are very similar, and add coherently, whereas their intrinsic noise signals are incoherent.
In a compact array of a large number of small MEMS microphones, the microphones generally have high intrinsic noise levels. Furthermore, because compact arrays are not very sensitive to low-frequency sound, their low-frequency performance is degraded by high levels of microphone intrinsic noise. Summing the outputs of a large number of adjacent microphones is thus especially beneficial for compact MEMS arrays. It will be understood that although the invention derives particular benefit when the microphones such as MEMS are fabricated with the switching and processing circuitry (e.g., in a single integrated circuit), the switching and processing circuitry may of course be fabricated separately from the microphones and connected subsequently thereto.
It will be understood that other alternatives to the embodiments of the invention described above will be apparent to a person of ordinary skill in the art .