CN108702143B

CN108702143B - Digital filter with confidence input

Info

Publication number: CN108702143B
Application number: CN201780013208.9A
Authority: CN
Inventors: A·海姆; M·霍赫
Original assignee: Microchip Technology Germany GmbH
Current assignee: Microchip Technology Germany GmbH
Priority date: 2016-07-21
Filing date: 2017-07-18
Publication date: 2023-08-01
Anticipated expiration: 2037-07-18
Also published as: EP3488524A1; TW201806316A; KR102431556B1; JP2019525502A; TWI721194B; JP6913099B2; CN108702143A; KR20190031429A; WO2018015412A1

Abstract

The invention discloses a digital filter, which comprises: an assigned filter function having assigned filter coefficients (b _k ) The method comprises the steps of carrying out a first treatment on the surface of the An input receiving an input sample (x _k ) The method comprises the steps of carrying out a first treatment on the surface of the Another input receiving a confidence value (c _k ) The method comprises the steps of carrying out a first treatment on the surface of the An output (out). Each input sample value (x _k ) And input confidence value (c) _k ) And wherein the filter output (out) is dependent on the input samples, the input confidence value, and the filter coefficients. The filter contains a plurality of accumulators (acc 01, acc02, acc03, acc 04), wherein after a predetermined number of sample values with associated confidence values have been input to the filter, output samples are generated.

Description

Digital filter with confidence input

Technical Field

The present invention relates to digital filters, and in particular to digital filters for noise suppression.

Background

To sense analog signals for processing in digital devices, sampling the signals faster (significantly) than the actual information content of the signals is a convention that allows the digitized signals to be enhanced to take full advantage of the redundancy of the information. Examples for such devices include capacitive-touch sensing or touchless position and gesture sensing systems, digital voltmeters, thermometers, or pressure sensors.

AN exemplary capacitive sensing system that can suffer from severe noise includes "mTouch" in application note AN1478 ^TM Sensing solution acquisition method capacitive voltage divider (mTouch ^TM Sensing Solution Acquisition Methods Capacitive Voltage Divider) "microchip CTMU (Microchip CTMU for Capacitive Touch Applications)" for capacitive touch applications in AN1250 "The system described, both annotations from the assignee of the present application, microchip technology company (Microchip Technology Inc), and incorporated herein by reference in its entirety.

Another exemplary application is a touchless capacitive 3D gesture system, also known asTechniques.

The sensor signal typically experiences interference caused by various noise types, such as wideband noise, harmonic noise, and peak noise. Harmonic noise and peak noise may be due to, for example, switching power sources and are also addressed in electromagnetic immunity standard tests (e.g., IEC 61000-4-4).

Signal acquisition may also be interrupted in a predetermined or deterministic scheme, for example, when some sensors are multiplexed in time or due to irregular events (e.g., data transmission failure). Such an interruption or missing sample may cause an undesirable phase jump in the signal. Since digital filters are designed for regular sampling intervals, this will damage the filter timing and can severely impact its noise suppression performance.

Similar to erased messages in the context of channel coding for digital communications (Blahut), 1983, bossert, 1999, we refer to missing samples and samples that do not carry useful information (e.g., due to peak noise) as erasures.

Fig. 1a shows a system 100 that performs the basic procedure for estimating a noise real-valued baseband signal. Analog-to-digital converter (ADC) 110 samples the signal at a rate (significantly) higher than the information change thereof. The digital signal is then input to the low pass filter 120 and downsampled by the integer multiple downsampler 130 with an integer multiple of the rate R. The downsampled result is further processed or simply displayed on a digital display 140, such as shown in fig. 1 a. Wherein the low pass filter 120 is able to attenuate the higher frequency components of the wideband noise but will not completely suppress the noise peaks.

Problems with peak noise suppression occur in many applications, such as image processing (t. Bei Naji mol (t.benazir), 2013), seismology and medicine (b.bosachash, 2004). The standard practice for combating peak noise is to apply a median filter or variant.

The approach to suppress peak noise but still smooth the input signal is to take the average of a subset of samples in the time window (excluding samples that have been identified as noise peaks or outliers or excluding, for example, n largest and n smallest samples (selective arithmetic mean (SAM) filters or 'Sigma (Sigma) filters' (Lee, 1983)). Obviously, SAM filters are time-variable filters that have a Finite Impulse Response (FIR) that adapts to the time domain characteristics of their input signal.

However, although the presence of noise peaks (i.e. with erasures) is preferred, without peaks the noise suppression characteristics of this SAM averaging filter are lower than, for example, filters using Hamming (Hamming) windows as the most advanced technique for impulse response or filters using the least squares method (as shown in fig. 1 b) for window lengths of 32 samples to design the frequency response. The solid curve of the least squares filter and the dashed curve of the hamming filter show improved sidelobe reduction compared to the average filter with rectangular impulse response (dot-dashed curve) in terms of the magnitude response of the filter.

Disclosure of Invention

There is a need for improved methods and systems for processing signals that are subject to noise. The present application is not limited to any of the sensor systems described above but is applicable to any type of signal that is subject to noise and needs to be evaluated.

According to an embodiment, a digital filter may include: an assigned filter function having assigned filter coefficients; an input that receives an input sample; another input that receives a confidence value; and an output, wherein each input sample value is associated with an input confidence value and wherein each input sample is weighted with its associated confidence value, wherein the filter output is dependent on both the input sample and the input confidence value, and wherein the filter comprises an accumulator configured to accumulate a predetermined number of confidence weighted input samples, the associated confidence value, confidence values weighted with assigned filter coefficients, and confidence weighted further with the assigned filter coefficients.

According to a further embodiment, the filter may comprise: a first branch having a first accumulator receiving input confidence values weighted with coefficients from a set of coefficients and generating a first accumulated value; a second branch having a second accumulator that receives the input confidence value and generates a second accumulated value; a third branch having a third accumulator that receives input sample values weighted with coefficients from the set of coefficients and the input confidence value and generates a third accumulated value; and a fourth branch having a fourth accumulator that receives the confidence weighted input value and generates a fourth accumulated value. According to a further embodiment, the first accumulated value is subtracted from a fixed value, wherein a result of the subtracting is divided by the second accumulated value and multiplied by the fourth accumulated value and added to the third accumulated value, and wherein the first accumulated value, the second accumulated value, the third accumulated value and the fourth accumulated value are subsequently cleared.

According to another embodiment, a digital filter may comprise: an assigned filter function having a first set of filter coefficients and a second set of filter coefficients; an input that receives an input sample; another input that receives a confidence value; and an output, wherein each input sample value is associated with an input confidence value; the filter output is dependent on both the input samples and the input confidence value, and wherein the digital filter comprises: a first branch having a first accumulator receiving the input confidence value weighted with coefficients from the first set of coefficients and generating a first accumulated value; a second branch having a second accumulator that receives the input confidence value weighted with coefficients from the second set of coefficients and generates a second accumulated value; a third branch having a third accumulator that receives input sample values weighted with coefficients from the first set of coefficients and the input confidence value and generates a third accumulated value; and a fourth branch having a fourth accumulator that receives an input value weighted with coefficients from the second set of coefficients and the input confidence value and generates a fourth accumulated value.

According to a further embodiment of the above digital filter, the first accumulated value is subtracted from a fixed value, wherein a result of the subtracting is divided by the second accumulated value and multiplied by the fourth accumulated value and added to the third accumulated value, and wherein the first, second, third and fourth accumulated values are subsequently cleared.

According to a further embodiment of any of the above digital filters, multiple examples of filters are operated in parallel, each example involving a subset of input samples and associated confidence values and having proprietary coefficients. According to a further embodiment of any of the above digital filters, the confidence value is represented by a digital logic value. According to a further embodiment of any of the above digital filters, the fixed value is the sum of all coefficients. According to a further embodiment of any of the above digital filters, the assigned filter function is a low pass filter function. According to a further embodiment of any of the above digital filters, the low pass has been obtained from converting a high pass or band pass into an equivalent low pass domain. According to a further embodiment of any of the above digital filters, the assigned filter function has only positive coefficients or only negative coefficients. According to a further embodiment of any of the above digital filters, the assigned filter function has at least one non-zero value coefficient having a different magnitude than another non-zero coefficient. According to a further embodiment of any of the above digital filters, the DC gain of the digital filter is constant or substantially constant.

According to another embodiment, a filter system may comprise: a first digital filter and a second digital filter each comprising an assigned filter function having assigned filter coefficients, an input receiving input samples, another input receiving confidence values, and an output, wherein each input sample value is associated with an input confidence value and wherein each input sample is weighted with its associated confidence value; wherein the filter output is dependent on both the input samples and the input confidence values, and wherein the filter comprises an accumulator configured to accumulate the confidence weighted input samples, the associated confidence values, the confidence values weighted with assigned filter coefficients, and the confidence weighted input samples further weighted with the assigned filter coefficients; and a demultiplexer that receives an input signal and generates input samples for the first digital filter and the second digital filter.

According to a further embodiment of the filter system, the system may further comprise: a first outlier detector that receives the input samples for the first digital filter and generates an associated confidence value; and a second outlier detector that receives the input samples for the second digital filter and generates an associated confidence value. According to a further embodiment of the filter system, the input samples for the first digital filter are high samples and the input samples for the second digital filter are low samples.

According to a further embodiment, a filter system may comprise a first digital filter and a second digital filter each comprising a distributed filter function having a first set of filter coefficients and a second set of filter coefficients, an input receiving input samples, another input receiving confidence values, and an output, wherein each input sample value is associated with an input confidence value; the filter output is dependent on both the input samples and the input confidence values, and wherein each of the digital filters comprises: a first branch having a first accumulator receiving the input confidence value weighted with coefficients from the first set of coefficients and generating a first accumulated value; a second branch having a second accumulator that receives the input confidence value weighted with coefficients from the second set of coefficients and generates a second accumulated value; a third branch having a third accumulator that receives input sample values weighted with coefficients from the first set of coefficients and the input confidence value and generates a third accumulated value; and a fourth branch having a fourth accumulator receiving the input value and the input confidence value weighted with coefficients from the second set of coefficients and generating a fourth accumulated value; and wherein the system further comprises a demultiplexer that receives an input signal and generates input samples for the first digital filter and the second digital filter.

According to a further embodiment of the above filter system, the system may further comprise: a first outlier detector that receives the input samples for the first digital filter and generates an associated confidence value; and a second outlier detector that receives the input samples for the second digital filter and generates an associated confidence value. According to a further embodiment of the above filter system, the input samples for the first digital filter are high samples and the input samples for the second digital filter are low samples.

According to another embodiment, a digital filter may comprise: an assigned filter function having assigned filter coefficients; an input that receives an input sample; another input that receives a confidence value; and an output, wherein each input sample value is associated with an input confidence value; wherein the filter output is dependent on the input samples, the input confidence value, and the filter coefficients; wherein the filter contains a plurality of accumulators; wherein the output samples are generated after a predetermined number of sample values, wherein the associated confidence values have been input to the filter.

According to a further embodiment, a method for filtering digital input samples may comprise the steps of: receiving a digital input sample value and an associated input confidence value; accumulating the input confidence values weighted with coefficients from a set of coefficients and generating a first accumulated value; accumulating the input confidence values and generating a second accumulated value; accumulating the input sample values and the input confidence values weighted with coefficients from the set of coefficients and generating a third accumulated value; the confidence weighted input values are accumulated and a fourth accumulated value is generated.

According to a further embodiment of the method, the method may further comprise: the first accumulated value is subtracted from a fixed value, wherein a result of the subtracting is divided by the second accumulated value and multiplied by the fourth accumulated value and added to the third accumulated value, and then the first accumulated value, the second accumulated value, the third accumulated value, and the fourth accumulated value are cleared. According to a further embodiment of the method, the fixed value is the sum of all coefficients. According to a further embodiment of the method, the input confidence value is binary.

According to a further embodiment, a method for filtering digital input samples may comprise the steps of: receiving a digital input sample value and an associated input confidence value; accumulating the input confidence values weighted with coefficients from a first set of coefficients and generating a first accumulated value; accumulating the input confidence values weighted with coefficients from a second set of coefficients and generating a second accumulated value; accumulating input sample values weighted with coefficients from the first set of coefficients and the input confidence values and generating a third accumulated value; and accumulate the input value and the input confidence value weighted with coefficients from the second set of coefficients and generate a fourth accumulated value.

According to a further embodiment of the method, the method may further comprise subtracting the first accumulated value from a fixed value, wherein a result of the subtracting is divided by the second accumulated value and multiplied by the fourth accumulated value and added to the third accumulated value and subsequently clearing the first, second, third and fourth accumulated values. According to a further embodiment of the method, the fixed value is the sum of all coefficients. According to a further embodiment of the method, the input confidence value is binary.

Drawings

FIG. 1a shows an analog signal and exemplary acquisition of analog-to-digital conversion and conventional noise suppression;

FIG. 1b shows magnitude response of different low pass filters;

FIG. 2 shows typical tap weights for a low pass filter with finite impulse response;

FIG. 3 shows an exemplary block diagram of a data source and related confidence generation as an input source for a digital filter;

FIG. 4 shows an exemplary implementation of a digital filter with confidence input;

FIG. 5 shows a system with an external confidence generation controller;

FIG. 6 shows an exemplary shift register implementation of a digital filter with confidence inputs;

FIG. 7 shows an example of redistribution of erased coefficient weights, in accordance with various embodiments;

FIG. 7A shows another example of redistribution of erased coefficient weights in accordance with various embodiments of a two-level signal;

FIG. 8 shows an example of peak noise suppression performance;

FIG. 9 shows a comparison of filter coefficients with magnitude spectra with and without erasures;

FIG. 10 shows an example of redistribution of erased coefficient weights in a high pass filter embodiment;

FIG. 11 shows an example of a non-touch gesture detection system using alternating quasi-static electric fields;

FIG. 12 shows an example of a two-level measurement;

FIG. 13 shows an implementation for packet data processing with minimal buffering requirements;

fig. 14 shows an implementation similar to fig. 13 but utilizing a converter to implement multiplication with binary confidence ck=0, 1;

FIG. 15 shows an implementation similar to FIG. 13 but with differential stages after integer multiple downsampling;

FIG. 16 shows an implementation similar to FIG. 13 but with a generic redistribution function;

FIG. 17 shows an implementation of binary confidence input with generic redistribution function and implementation with switches;

FIG. 18 shows the original filter impulse response divided into two parts; and is also provided with

Figure 19 shows the use of an ADC output sample r with modulation for amplitude _k An embodiment of two filters for confidence inputs of (a).

Detailed Description

According to various embodiments, a real-valued baseband signal may be obtained (e.g., demodulated and downsampledSignal) where the input signal is sampled and noisy. />A device (e.g., MGC3030 or MGC 3130) or updated design is available from the assignee of the present application. For example, FIG. 11 shows an exemplary embodiment in which controller 740 represents +.>And (3) a device. For example, 2015, 1, 15, on-line publication of->Design guidelines (/ ->Design Guide) "overview and Design guidelines are obtained from microchip technology corporation and are incorporated herein by reference.

The 3D gesture detection system 700 shown in fig. 11 provides a transmit electrode 720 and a plurality of receive electrodes 710a..d that may be formed from a frame structure as shown in fig. 11. However, the entire rectangular area under the receiving electrode 710a..d may be used as the transmitting electrode, or this electrode may also be divided into a plurality of transmitting electrodes. The emitter electrode 720 generates an alternating electric field. Gesture controller 740 receives signals from receive electrodes 710a..d that may represent capacitances between receive electrodes 710a..d and system ground and/or transmit electrode 720. Gesture controller 740 may evaluate the signals and provide human device input information to processing system 730. This information may be 3D movement coordinates similar to 2D movement information generated by a computer mouse and/or include commands generated from detected gestures.

The problem faced in this application is that the noise introduced into the sensor signal is broadband and peak noise, which is not known to be solved simultaneously by any current state-of-the-art practice. In addition, inIn applications and other applications, some samples of the input signal may have been lost or may not be available for various reasons. Although the negative effects of input noise are evident, irregularities in the input sample interval lead to corruption of the filter timing and severely impact noise suppression performance. Digital filters are typically designed for regular sampling intervals, but in addition to this they lead to undesirable phase jumps in the input signal from the filter's perspective. The location of noise peaks and missing samples in the signal is determined by some other means, such as a peak noise detection system or deterministic noise indicator. As mentioned above, there is a standard practice for combating wideband noise, namely applying a frequency (low pass) filter. Furthermore, there is another standard practice for combating peak noise, namely to apply a median filter within a window of signal samples.

Although such problems areIs particularly relevant in the system (as mentioned above), but it is not possible to apply such cases only to +. >The system, these cases may also be related to other sensor systems. Thus, the proposed measures are applicable to various signal sources.

For the proposed filtering method, each input sample is associated with a confidence value. This confidence value indicates whether the associated sample is erased-i.e., whether it is indeed a missing sample or a sample known not to carry useful information. It is assumed that some other component knows the confidence value. Such means may include, for example, deterministic inputs, or outlier detection methods such as the glaubes test (Grubbs, 1950), the Generalized Extremum Student's Dispersion (GESD) test, or the hapel identifier (hapel, 1974). In the context of image processing, the confidence value is used as a weight (j. Hao Renchuan p (j.horentrup), 2014) in a least squares regression for improved alpha matting, for example.

According to various embodiments, the following should be adhered to suppress wideband noise and ignore unwanted noisy or missing samples, for example, it is known that none of the current state-of-the-art practices can solve both of the problems at the same time:

1. the erasure (e.g., detected noise peaks) must not contribute to the filter output;

2. Constant filter gain should be provided at DC (for a constant input signal, the filter output signal level must also be constant);

3. gradual adaptation to the number of erasures should be provided while preserving default filter characteristics when no erasures are present

Fig. 2 shows the filter coefficients of a typical low-pass filtering (or 'windowing') function, here exemplified by a normalized hamming window of length 8. Each filter coefficient defines the weight of its associated tap in a tapped delay line implementation, also shown in fig. 2, with each tap aligned below its associated coefficient in the histogram. Thus, we use the terms 'filter coefficients' and 'tap weights' synonymously. In this example, the tapped delay line contains 7 consecutive delay stages z ^-1 8 tap weights. An exemplary input sample is also shown in fig. 2. Other sample structures may have fewer or more stages.

According to various embodiments, in the case of a filtering function and an input signal with confidence information per sample, the weight of the filter tap corresponding to the input sample with less confidence is reduced while maintaining the DC gain of the filter. Fig. 2 shows that in this example the delay stage z is delayed by, for example, 7 consecutive stages of delay z ^-1 Typical weight/coefficient distribution of the resulting low pass filter. Other sample structures may have fewer or more stages.

In the following input samples that do not carry any useful information, equivalently missing samples may be considered erasures, and corresponding samples are referred to as having zero confidence. It is assumed that the information of whether the sample is erased is known (e.g., by comparing the sample to a threshold) from any other source or algorithm. The impulse response of the corresponding digital filter will be referred to as the 'filter function'.

Fig. 3 shows a block diagram of an exemplary digital filter 300 with confidence inputs and their input signal sources. The data source 320 generates a sample having samples x at discrete times k _k Is a signal x of (a). The signal x is input to the peak noise (or 'outlier') detector 330, and the peak noise (or 'outlier') detector 330 is determined by determining the confidence value c _k And x _k Correlation will be per sample x _k Classified as 'no noise' or 'noisy', where e.g. c _k =1 means ' without noise ' or ' x _k Full confidence', and c _k =0 means ' noisy ' or ' x _k Without confidence. I.e. with associated c _k Sample x=0 _k Is the wiping off. According to other embodiments, the confidence information may also originate from some external component called external indicator 310. This external indication may be considered a deterministic confidence input.

FIG. 5 shows a system 500 with a 2D capacitive touch detection and finger tracking system, as used in a touchpad or touch display, for example, the system 500 is comprised of a number of sensor electrodes "2D electrode patterns" 520 and a controller unit "2D touch controller" 510. Four further electrodes A, B, C, D used in conjunction with the 3D gesture controller 530 to form a capacitive 3D gesture detection system are arranged around the 2D electrode pattern. When the 2D touch detection system 510, 520 is active, it interferes with the received signal of the 3D gesture detection system, i.e., the received data of the 3D gesture detection system is noisy and unusable. To generate a functional 2D-3D capacitive sensor system 500,2D touch controller 510 is only occasionally active when no touch is detected, and when the 2D touch controller 510 is active, this is signaled to the 3D gesture controller 530 (dashed arrow), the 3D gesture controller 530 then knows that its current reception value is noisy and with zero confidenceThe degrees are associated. That is, when sample x is generated in the event that the 2D system 510, 520 interferes with the received signal _k The external indicator may be set to c _k =0, and otherwise c _k =1. Will x _k C _k Is input to a digital filter with confidence inputs within the 3D gesture controller 530.

Examples for simple peak noise detectors or outlier detectors are as follows: when at each time k, the last M samples x are calculated _k-1 ,x _k-2 ,…,x _k-M Average value of (2)Standard deviation ofIf |x _k |>μ _k +3·σ _k Then set c _k =0, otherwise c _k ＝1。

1. The main method is that

An N-order standard digital Finite Input Response (FIR) filter is considered to have a time invariant filter function b= [ b ] ₀ ；b ₁ ；…；b _N ]Wherein b _i I=0, 1, …, N is a filter coefficient. For having sample x _k Given input signal x of (2), the filter output signal y is

(1.1)

Where k is a discrete time index. All filter coefficients b _i Is the Direct Current (DC) filter gain. For simplicity and without loss of generality, we assume the following

For each input sample x _k We assume we have an associated confidence value c _k . Initially we assume that the confidence value is binary, where c _k E {0;1}; wherein c _k By =0 is meant' sample x _k Internal no confidence', and c _k =1 means 'full confidence'. From having coefficient b _i In the time-invariant filter function b of (2), we will calculate the coefficient w _i (k) A time-variable filter function w (k), coefficient w _i (k) Depending on the confidence value c _k . C when the last n+1 input samples have all had full confidence, i.e. for i=0, …, N _k-i =1, we expect the filter function w (k) to be equal to the function b. However, if there is one or more erasures (i.e., with associated c _k-i Input sample x=0 _k-i ) Next x _k-i Must not contribute to the output value y _k 。

This is accomplished by combining each of the filter coefficients b in (1.1) _i Multiplying its associated input sample x _k-i Confidence value c of (2) _k-i And is realized. However, due to the modified filter coefficient b _i ′(k)：＝b _i ·c _k-i So that the DC filter gain is no longer guaranteedIs constant.

Therefore, the erased filter weights must be distributed over the other filter coefficients. The preferred way to do this is to distribute the weights of the erasures

Evenly distributed in the restWill produce a non-erasure coefficient with coefficient w _i (k) Linear Time Variable (LTV) filter

(1.2)

Where i ε {0,1, …, N }, and we denoteAs a sample x at a time k _k-i Associated relative confidence.

Such is the same as substituting the erased input samples with the average of the unerased input samples at each time instance k, and setting c for all i _k-i =1. This way of implementing the algorithm is of great interest for a one-time or 'block-wise' processing of each input sample, just as the algorithm is done with windowing and DC value operations to estimate the DC values of a finite set of consecutive samples.

Proof of evidence

Since this procedure means that the input data is over-written, it is not suitable for continuous filtering, where each input sample results in multiple output samples, and the calculation of the output value is done at a higher rate than the input rate divided by the filter length, where we define the filter length as (n+1), i.e. the filter order plus 1.

The redistribution of erased filter coefficient weights is visualized in fig. 7. At the top, the last 8 input samples, x, at a known time k _k-4 X is a group _k-1 Is the erasure of the 8 input samples. The first plot shows the coefficients b of the initial filter (length 8 hamming window) aligned at the bottom with the shift register implementation _i . In the second graph, when the corresponding input sample x _k-4 X is a group _k-1 When it is erased at time k, coefficient b ₁ ' (k) and b ₄ The value of' (k) is set to zero. The sum of erased coefficients is also shown on the far right side of the second plot. In the third plot, the erased weights as shown in the right portion of the second plot are evenly redistributed over the coefficients assigned to the unerased input samples, resulting in w _i (k) A. The invention relates to a method for producing a fibre-reinforced plastic composite The added weights are shown to be different from the weights planned in the third graph. In this embodiment, the coefficient weights are shown in the bottom shift register filter diagram Is w ₀ To w ₇ 。

For the next input sample at time k+1, the sample and its corresponding confidence information are moved to the right within the shift register of the filter and for generating a different filter coefficient w _i The erased shift pattern of (k+1) again completes the redistribution.

An example of the noise suppression performance of the filter is shown in fig. 8. The top plot shows the filter input signal, which is a slowly varying information signal with added Gaussian (Gaussian) noise, some noise peaks starting at the sample index 250, and added 60Hz sinusoidal noise. The second graph shows that conventional low pass filtering with a hamming function of length 64 reduces higher frequency noise but smears only noise peaks that occur in the input signal. However, after the noise peaks have been identified, they are completely suppressed according to various embodiments. The bottom graph of fig. 8 demonstrates the advantage of using a hamming erasure filter function instead of a moving average: hamming erasure filtering results in better suppression of wideband noise contained in the input signal, and thus smoother output, than simply taking the average of the non-peak samples (i.e., selective arithmetic average filtering).

Fig. 9 illustrates how the erasure affects the frequency response of the filter. Here, the left side shows a typical low pass filter using rectangular and hamming windows and their associated magnitude spectra. The same filtering using two erased samples is shown on the right. The spectrum of a hamming erasure filter can be found to resemble a rectangular erasure filter, depending on the location of the erasure.

2. Summarizing

2a) Non-binary confidence input

Up to this point, the confidence inputs are binary, i.e., the associated input samples may or may not be used to compute the filter output values. However, confidence inputs can be summarized directly from the above comments to obtain a real value between 0 and 1 (i.e., c _k ∈[0,1]) And c _k The larger we confidence the associated sample x _k . Except for c _k Equation (1.2) remains the same.

2b) Generic redistribution function

For binary confidence inputs, in equation (1.2), the erased weights are evenly distributed over the other coefficients. Two terms in (1.2) can be interpreted as outputting two parallel filter branches that have been summed. The filter function in the first term is computed from b and the confidence input, and the second term is a time-variable average filter. By introducing a further FIR filter function g and coefficients g _i Summarizing the second term to produce

This is also applicable to non-binary confidence input c _k ∈[0,1]. We are marked as

Weighting g relative confidence with sample x at time k _k-i And (5) associating.

Thus, the filter output is given by

A possible implementation of this filter is shown in fig. 4, where the block denoted 'B' refers to a standard FIR filter with a filter function B, and similarly the block denoted 'G' refers to a standard FIR filter with a filter function G, and denoted asThe block of (a) refers to 1 divided by the input data of the block, i.e., the output of this block is the multiplicative inverse (reciprocal) of the input. In this embodiment, fourThe filter coefficients of the filter blocks ('B' and 'G') are constant. Of course, processing the same input data (i.e., c _k Or q _k ：＝x _k ·c _k ) The filters 'B' and 'G' of (a) may share the same buffer, as shown for the n=7 th order filter in the shift register implementation in fig. 6, highlighting weighting the input data value q for confidence _k ：＝x _k ·c _k Confidence value c _k Delay lines for both. Here, the input signal of the filter block contains an adaptation. At the same time, the implementation is equivalent to having the appropriate filter coefficients w _i (k) Is a single FIR filter of (c).

The characteristic nature of the tap-delay line implementation of the FIR filter in fig. 6 is that the tap-delay line for confidence value TDL-C and the tap-delay line for confidence weighted input data TDL-XC are identical, i.e. they have the same number of delay stages, and the same tap weight b ₀ ,b ₁ … and g ₀ ,g ₁ … are connected to the respective delay stages. Of course, one or the other delay line may be simplified depending on the type of input variable (e.g., binary confidence input). In addition, when g ₀ ＝g ₁ ＝g ₂ When= …, the delay line or associated operation block can be simplified. Furthermore, if weight b in TDL-C ₀ ,b ₁ … differs from the weight b in TDL-XC in fixed factor ₀ ,b ₁ Weight g in … or TDL-C ₀ ,g ₁ … differs from the weight g in TDL-XC in a fixed factor ₀ ,g ₁ … is insignificant because such factors can compensate for the outside of the tapped delay line.

For example, if g _i =1/8, then the corresponding tap weight reset position may also be 1, so the multiplication is preserved, and only the sum at the end of the tap delay line before the (1/x) block is divided by 8, which may also be done by a bit shift operation.

Confidence weighting input data values q implemented by multiplication in fig. 6 for binary confidence inputs or confidence values from a finite set of values _k The operation of (a) may also be performed by conditional statements (e.g., IF/ELSE or SWITCH statement) implementation, e.g., if c _k =0, q _k Set to 0, and if c _k =1, then q _k Set to 1. The conditional statement may also be specified at each tap of the delay line compared to before the delay line: then, if associated c _k-i Is 1, then only tap-weighted input value b _i ·x _k-i Or g _i ·x _k-i Added to the sum of the outputs of the respective delay lines. In this case, sample x may be taken _k Directly to TDL-XC without prior multiplication by c _k . The same is true for TDL-C.

The order of the filters with impulse responses b and g need not be equal. Without loss of generality, the filter is defined to have an equal order N, where N is at least as large as the maximum of the orders of the filter with b and g, and the unused coefficients are assumed to be zero.

Selecting g=b yields another non-preferred approach for redistributing the erased coefficient weights. The unerased filter coefficients are scaled by the same factor, which is recalculated at each discrete time instance k, i.e

3. Exception handling

With respect to the denominator in (1.2) or (1.3), the output value is obviously not operable if the last n+1 input samples all have zero confidence. A possible exception to this is to repeat the last valid output sample, or to forward the exception to a subsequent processing stage.

4. Windowing and DC value operation, particularly for signals having two or more desired signal levels

For symmetric filtering or "windowing" function b, taking a snapshot when the input signal is convolved with function b is equivalent to weighting the input samples with b and summing the point-wise products. Thus, the above concepts can be equally applied when the DC value of the windowed signal is of interest. The main difference between windowing and DC operation and continuous filtering is that for windowing and DC operation typically only a single output value is caused per input sample, i.e. this is one-time processing or block-wise processing of the input samples.

In many applications, the measurement signal, which typically has additional noise, alternates between two different levels. We refer to these levels as 'high' signal levels and 'low' signal levels. FIG. 12 shows exemplary measurements with such high and low levels. An example is Amplitude Modulation (AM) with synchronously sampling an analog received signal at twice the carrier frequency, where the information is accommodated in the difference between 'high' signal level and 'low' signal level. This method is applied, for example, in a capacitive touch detection system orIn the art. The measured (or 'received') signal of such an AM sensor system can be demodulated, for example, by alternately multiplying +1 and-1, and then low pass filtered to estimate the DC (zero frequency) value-the actual information, i.e., 'average' difference between 'high' samples and 'low' samples.

This signal having two levels will now be considered, which is input to a low pass filter in standard applications, where the system nomenclature of 'high' samples and 'low' samples keeps labeling the set of samples corresponding to one of two different signal levels. It will be assumed that deviations in their respective signal levels are caused by noise.

When it is detected that a 'low' sample is useless, for example due to detected peak noise, we zero the weighted position of its corresponding coefficient in the filter and redistribute the erased weights over other coefficients. However, in order to maintain the expected value of the filter output, it must be redistributed only over coefficients assigned to other 'low' samples. Otherwise, the filter output will be closer to a 'high' level than would otherwise be the case, as the coefficients assigned to the 'high' samples will get additional weight.

In general, samples of an input signal must be classified into sets of samples having the same expected value, and digital filtering with confidence inputs (i.e., redistribution of coefficient weights) must occur in such a way that the ownership weights assigned to the coefficients of each set remain constant, which is most easily accomplished when redistributing erased weights in one set within the same set.

FIG. 7A shows an example windowing and DC operation of a signal having two levels. The processing of erased values is performed separately for samples at 'high' and 'low' signal levels, respectively, as shown in fig. 7A. Every other filter coefficient is assigned to a measurement in either a 'high' signal level or a 'low' signal level. Graph a) shows the initial filter coefficients in fig. 7A. Graph b) separates 'high' level coefficients and graph c) separates 'low' level coefficients. Chart d) shows in fig. 7A that coefficient 3 corresponds to a sample at a 'low' signal level and erases. Its weights are redistributed over other coefficients assigned to samples with 'low' signal levels. Diagram e) shows the redistribution of coefficients relative to 'low' signal levels. Bottom graph f) shows a combination of redistributed 'low' signal level coefficients and 'high' signal level coefficients. Thus preserving the expected output value. This method may be implemented with two data branches, one for samples with 'high' signal levels and the other for samples with 'low' signal levels, with the outputs of the branches being summed last. Furthermore, since this is a time-variable filter, the redistribution of the filter coefficients is updated for each output sample.

5. Confidence output

The ability to immediately process the input confidence value creates a problem if confidence values can also be provided to each output sample. This output confidence measure should be independent of the input sample value, but should be a function of the input confidence value and the filter coefficients only, i.e

(1.4) d _k ＝f(b ₀ ,b ₁ ,…,b _N ,c _k ,c _k-1 ,…,c _k-M )

For a positive integer M.

The measurement that satisfies (1.4) and is readily available is the sum of the filter coefficients weighted with their corresponding input confidence values, i.e

Support d for this too _k ∈[0,1]So long as for all i, b _i Not less than 0 andand in particular d when all input samples have full confidence _k =1, and when all input confidence values are c _k-i When=0, d _k ＝0。

With this confidence input, a plurality of the mentioned filters can be cascaded. In addition, this output may be used for high level control, e.g. "if the output confidence is too low, then no touch event is triggered".

7. Design rule

The mentioned approach is applicable to any low-pass FIR filter. However, all filter coefficients should have the same sign, e.g. positive. Mainly, the requirement for constant DC filter gain can be met also when (part of) the tap weights are negative. However, this would introduce the hazard of producing undesirable filtering characteristics (e.g., high pass characteristics) for some confidence input clusters.

Furthermore, the more similar the filter coefficient values, the less important it is to erase the input samples assigned to coefficients with larger values. In particular, coefficients for rectangular windows, triangular windows, hamming windows, and Han En (Hann) windows obey these rules.

There are no further parameters to consider, other than the selection of initial filter coefficients and exception handling.

The approach can be extended to high pass filters. Fig. 10 shows an example of a high pass filter with initial filter coefficient weights alternating between positive and negative signs (as shown in the top first diagram). Thus, according to an embodiment, first the high pass filter coefficients are demodulated using alternating signs, as shown in the top second diagram. Then, a method of low pass filtering as shown in fig. 2 is applied, as shown in the top third and fourth graphs. The modified weights shown in the top fourth diagram are then remodulated using the inverse alternating signs. This results in distributed weights as shown in the top fifth graph. The input signal of the filter may also be demodulated as compared to re-modulating the filter coefficients and filtered with an equivalent low pass having coefficients according to the top fourth diagram before re-modulating the signal. Whether the input signal is filtered directly with a high pass filter or demodulated and filtered with an equivalent low pass filter, it is important that the samples of the input signal of the high pass filter (if it is to be demodulated) will have a single expected value.

8. Application & use case

As mentioned above, the proposed concept is applicable to any filtering system where the sampled input signal changes faster than the actual information (the higher the sampling rate the better). In many other systems, such systems include 3D capacitive sensor systems, such as the assignee GestIC system and 1D/2D capacitive touch solutions. The filtering method may be further applied to other sensor signals and is not limited to capacitive sensor systems.

9. Properties of (C)

When the filter is initialized with arbitrary but zero confidence data, it provides an estimate of the input signal from the start-up. Thus, when filtering a signal with a non-zero average value, the filter does not exhibit a typical step response and the filter conditions have been initialized to zero.

Digital examples

In the following we give a digital example for computing the output value of a filter with confidence input. The following table states the input sample value x at time k _k Associated confidence value c _k And the coefficient b of the initial filter function b _i . Here g is a fixed value, i.e. the erased coefficient weights will be evenly redistributed.

k	-7	-6	-5	-4	-3	-2	-1	0	1	2	3	4	5	6	7	8	9	10
																			x _k	0	0	0	0	0	0	0	7	6	8	9	6	21	6	7	33	8	6
c _k	0	0	0	0	0	0	0	1	1	1	1	1	0	1	1	0	1	1

i	0	1	2	3	4	5	6	7
									b _i	0.0207	0.0656	0.1664	0.2473	0.2473	0.1664	0.0656	0.0207
g _i	0.125	0.125	0.125	0.125	0.125	0.125	0.125	0.125

There is a erasure at times k=5 and k=8, i.e. c ₅ ＝c ₈ =0. For example, the treatment confidence has been set to zero here because the value x has been detected ₅ X is a group ₈ Is the noise peak.

For initialization of the filter, the first sample confidence pair (x) is entered (e.g., by entering N samples with zero confidence ₀ ,c ₀ ) All confidence information in the memory of the filter is previously set to zero. In the table, for k<0, then indicate c _k =0. At time k=0, there will be a confidence level c ₀ Sample x=1 ₀ =7 is input to the filter. According to the above equation, modified coefficient b _i ' k=0 and coefficient w _i (k=0) as

i	0	1	2	3	4	5	6	7
									b _i ′(0)	0.0207	0	0	0	0	0	0	0
w _i (0)	1	0	0	0	0	0	0	0

That is, x is ₀ =7 directly forwarded to output, i.e. y ₀ ＝x ₀ ＝7。

At time k=9, when two erasures x ₅ X is a group ₈ When in the buffer of the filter then b _i ' (k=9) and w _i (k=9) operation as

i	0	1	2	3	4	5	6	7
									b _i ′(9)	0.0207	0	0.1664	0.2473	0	0.1664	0.0656	0.0207
w _i (9)	0.0624	0	0.2081	0.2889	0	0.2081	0.1073	0.0624

Fig. 13-17 show other implementations of a erasure filter for packet data processing. These embodiments use different but very memory efficient solutions.

In packet data processing, each input sample results in only a single output value, and thus the implementation may be accomplished using the accumulator acc01..acc04 instead of a tapped delay line (buffer), as shown in the examples according to fig. 13-17. According to some embodiments, these accumulators are set to zero for each new data packet and then the corresponding input values are continuously added to the accumulators.

Furthermore, when the output data rate is lower than the input data rate due to some integer multiple downsampling factor, an intermediate version (plus accumulator) with a shorter delay line than in other embodiments is also possible.

Hereinafter, we denote the length of the data packet as L and the order of the low pass filter as N, where n=l-1. L samples of a data packet of length L are denoted as x _k K=0, 1, …, L-1. For each, this sample is associated with a binary confidence value, c _k ∈[0,1]I.e. 0.ltoreq.c _k Less than or equal to 1, wherein c _k =0 means, for example, a peakThe value noise or outlier detector considers x _k Does not carry any useful information and should not result in a filter output, and c _k =1 means x _k The filter output should be entirely caused.

Fig. 13 shows an implementation for packet data processing with minimal buffer requirements, in other words without any lengthy buffers, but with only four accumulators acc01, acc02, acc03 and acc04. For each data packet of length L to be processed, L input samples x _k K=0, 1..l-1 and its associated confidence level c _k Together into the filter. The corresponding filter coefficient b can be _k Stored in flash. According to some embodiments, after each data packet, it is necessary to reset accumulator acc0x to zero. It should be noted that the operation of lowering the right part of the sample block by an integer multiple labeled box "L ∈" is updated only once for each data packet. Thus, once the accumulator accumulates all of the input values for a packet having a length of L, the box "L ∈" only forms a gate that forwards the accumulator value at the input to its output. Thus, after accumulating the L values, the box "L ∈" will output the accumulated value. Fig. 13 also includes a label labeled () ^-1 I.e. the output value of the block is 1 divided by the input value.

For symmetric filter impulse response, where b at N for k=0, 1 _N-k ＝b _k In FIG. 13, b _N-k Can also be represented by b _k And (5) replacing.

For binary confidence input, where c _k E {0,1}, as compared to FIG. 14, the implementation of c with conditional increments of switches or variables may also be implemented _k I.e. weighted with confidence values.

According to other embodiments, if the accumulator is not reset after each packet L (which may not be the typical use case), the filter is no longer an FIR filter but may be an IIR filter.

In particular, "confidence weighted" data refers to the data being multiplied by a confidence value and using a switch to open or close a data path. Similarly, "weighting" and "weighted" refer to "multiplication" (multiplexing or multiplexing), respectively, but the multiplication may also occur by opening or closing the datapath.

Practical examples in the following a pseudo-code software implementation with symmetric filter impulse responses and binary confidence values is shown, according to fig. 14, using samples x_k and associated confidence values c_k and filter coefficients b_k to calculate the output of one data packet.

In contrast to employing accumulators acc01, acc02, acc03, and acc04 that need to be reset to zero after L samples have been input and final output values have been calculated, adding L input values may also be accomplished by recording the values of acc01, acc02, acc03, and acc04 prior to inputting the data of the current packet and subtracting these recorded values from the respective values of acc01, acc02, acc03, and acc04 after inputting the data of the current packet. This may be achieved, for example, by employing a first order CIC filter, which is an "economic digital filter for decimation and interpolation (An Economical Class of Digital Filters for Decimation and Interpolation)" as known, for example, from the institute of electrical and electronics engineers, acoustic, speech and signal processing journal (IEEE Transactions on Acoustics, spech, and Signal Processing) second chapter 155-162 of ASSP-29, entitled "economic digital filter for decimation and interpolation (Eugene b. Hogenauer)". This is shown in fig. 15, where after the accumulator and integer multiple downsampling stages, there is an additional differential stage of subtracting the previously accumulated and integer multiple downsampled values from the former.

Generic redistribution function

As for continuous data processing, the general redistribution function g may be employed for packet data processing as well. Next, fig. 13 and 14 are respectively fig. 16 and 17.

Sample value with multiple expected values

In many applications, such as capacitive sensing, the actual information is amplitude modulated and it is necessary to demodulate the ADC output data prior to low pass filtering, and both the ADC output value and the demodulated ADC output value have two desired values. However, a digital filter with confidence inputs may be applied directly to only input samples with a single expected value. Thus, the demodulated ADC output values need to be split into two sets, with a single expected value for all samples within each set.

Typically, the ADC alternately outputs samples (labeled low and high samples, respectively) at two different desired signal levels, and we assign an even time index k to the low samples and an odd time index to the high samples.

We assume symmetric filter impulse response, i.e. b _i ＝b _N-i I=0, 1,2,..n. We introduce x _k ^(L) ＝x _2k X is a group _k ^(H) ＝x _2k+1 And likewise for the filter coefficients, b _i ^(e) ＝b _2i And b _i ^(o) ＝b _2i+1 。

By using even i and odd i separation coefficients b _i While the initial low-pass filter impulse response is split into two parts. This is shown in fig. 18 for an example of an initial filter impulse response for a hamming window of length l=16.

By using these example filter impulse responses, fig. 19 shows how the demodulated samples x are distinguished for even and odd values _k And sample x _k How to split into two data branches, and where q is the bottom of k/2, according to FIG. 13, two examples of digital filters with confidence inputs but with different filter impulse responses (compare FIGS. 18 and 19) are used to process samples x with even and odd indices k, respectively _k . Each data branch in this example has its own peak noise detector that produces a confidence value c. The sample rate on each branch is the outputHalf the incoming sample rate and the packet length to be considered on each branch is the input sample x _k For example, where the initial packet length is l=16 and the packet length of each digital filter with confidence input is L' =l/2=8. The outputs of the two filters are added to produce the final result.

The digital filters mentioned above may be formed by hardware, such as a programmable logic device, or software in, for example, a microcontroller, a processor, or a digital signal processor.

Bibliography of

B. Baa Sha Shen, m.m. (B.Boashash, M.M.) (2004). Signal enhancement by time-frequency peak filtering (Signal Enhancement by Time-Frequency Peak Filtering). Signal processing Congress of the institute of Electrical and electronics Engineers (IEEE Transactions on Signals Processing), 929-937.

Brag Hu Te, r.e. (Blahut, r.e.) (1983). Theory and practice of error control codes (Theory and Practice of Error Control Codes). Redine, mass: addison-Wesley Press (Addison-Wesley).

Bosetet, m. (bosert, m.), 1999. Telecommunication channel coding (Channel Coding for Telecommunication). New York, NY, USA: john Wiley parent-child company (John Wiley & Sons, inc.).

The industrial engineers committee (Commission, i.e.), to which publication dates are not noted. Electric fast transient/burst immunity test (Electrical Fast Transient/Burst Immunity Test) (IEC 61000-4-4).

Glabuss, f.e. (Grubbs, f.e.) (3 months in 1950). Sample criteria for abnormal observations are tested (Sample criteria for testing outlying observations). Annual description of mathematics (Ann. Math. Static), pages 27-58.

H. Yellow, r.h. (h.hwang, r.h.) (1995). An adaptive median filter: new algorithms and results (Adaptive median filters: new algorithms and results). The society of electric and electronic engineers image processing journal (IEEE Transactions on Image Processing), 499-502.

Hanpel, F.R. (Hampel, F.R.) (1974). The curve and its role in robust estimation are affected (The influence curve and its role in robust estiamteion). Journal of the American society of statistics (Journal of the American Statistical Association), 69, 383-393.

J. Huo Lunte p, m.s (j. Horentrup, M, s.) (2014). Confidence-aware guided image filters (Confidence-aware guided image filter). The society of electric and electronic engineers International image processing conference (IEEE International Conference on Image Processing), (pages 3243-3247).

K. Ai Shiwa, v.j. (K.Aiswarya, V.J.) (2010). A new efficient algorithm (A new and efficient algorithm for the removal of high density salt and pepper noise in images and videos) for removing high density salt and pepper noise in images and video. Second computer modeling and simulation International conference (Second International Conference on Computer Modeling and Simulation), (pages 409-413).

Kalman, r.e. (Kalman, r.e.) (1960). A new method (A new approach to linear filtering and prediction problems) of linear filtering and prediction problems. Journal of basic engineering of the American society of mechanical Engineers (Transaction of ASME-Journal of Basic Engineering), 35-45.

Plums, j. -s. (Lee, j. -s.) (1983). A digital image smoothing and sigma filter (Digital image smoothing and the sigma filter). Computer Vision, graphics and image processing (Computer Vision, graphics, and Image Processing), 255-269.

Microchip touch and input sensing solutions (Microchip Touch and Input Sensing Solutions). (publication date is not noted). Retrieved from http:// www.microchip.com/touchhandinputsense

S.—j. Cox, y.h. (s.—j.ko, y.h.) (1991). A center weighted median filter and its application in image enhancement (Center weighted median filters and their applications to image enhancement). Electrical and electronic Engineers society of circuits and systems (IEEE Transactions on Circuits and Systems), 984-993.

T. Bei Naji, b.i. (t.benazir, b.i.) (2013). High-density and low-density impulse noise is removed from the digital image using a nonlinear filter (Removal of high and low density impulse noise from digital images using non linear filter). International conference on signal processing, image processing and pattern recognition (International Conference on Signal Processing, image Processing and Pattern Recognition), (pages 229-233).

Claims

1. A digital filter, comprising: an assigned filter function having assigned filter coefficients; an input configured to receive an input sample; another input configured to receive an input confidence value, wherein the input confidence value indicates a level of usefulness of an associated input sample; the output of the device is provided with a control unit,

Wherein each input sample value is associated with the input confidence value, and wherein the digital filter is configured to weight each input sample with its associated input confidence value to generate confidence weighted input samples;

the digital filter output is dependent on both the input samples and the input confidence value, and

wherein the digital filter comprises an accumulator configured to accumulate a predetermined number of the confidence weighted input samples, the associated input confidence values, the input confidence values weighted with assigned filter coefficients, and the confidence weighted input samples further weighted with the assigned filter coefficients, respectively.

2. The digital filter of claim 1, comprising:

a first branch having a first one of the accumulators configured to weight the input confidence value with the assigned filter coefficients and generate a first accumulated value;

a second branch having a second one of the accumulators, the second accumulator receiving the input confidence value and generating a second accumulated value;

A third branch having a third one of the accumulators configured to weight the input sample values with the assigned filter coefficients and the input confidence value and to generate a third accumulated value;

a fourth branch having a fourth one of the accumulators that receives the confidence weighted input samples and generates a fourth accumulated value.

3. The digital filter of claim 1, comprising:

a second branch having a second one of the accumulators configured to weight the input confidence value with coefficients from a second set of filter coefficients, and to generate a second accumulated value, wherein the second set of filter coefficients is for a generic redistribution function;

a third branch having a third one of the accumulators configured to weight the input sample values with the assigned filter coefficients and the input confidence values and to generate a third accumulated value;

A fourth branch having a fourth accumulator of the accumulators, the fourth accumulator configured to utilize coefficients from the second set of filter coefficients and the input sample values weighted by the input confidence values and to generate a fourth accumulated value.

4. The digital filter of claim 2, wherein the first accumulated value is subtracted from a fixed value, wherein a result of the subtraction is divided by the second accumulated value and multiplied by the fourth accumulated value, and a result of the multiplication is added to the third accumulated value, and wherein the first accumulated value, the second accumulated value, the third accumulated value, and the fourth accumulated value are subsequently cleared.

5. The digital filter of claim 1, wherein the confidence value is represented by a digital logic value.

6. The digital filter of claim 4, wherein the fixed value is a sum of all coefficients.

7. The digital filter of claim 1, wherein the assigned filter function is a low pass filter function.

8. The digital filter of claim 7, wherein the low pass has been obtained from converting a high pass or band pass into an equivalent low pass domain.

9. The digital filter of claim 1, wherein the assigned filter function has only positive coefficients, or only negative coefficients.

10. The digital filter of claim 1, wherein the assigned filter function has at least one non-zero value coefficient having a different magnitude than another non-zero coefficient.

11. The digital filter of claim 1, wherein the DC gain of the digital filter is configured to be constant or substantially constant.

12. The digital filter of any of the preceding claims 1-11, wherein the digital filter is formed by software.

13. A filter comprising a plurality of the digital filters of claim 1, wherein the plurality of digital filters operate in parallel, wherein each digital filter is configured to operate on a subset of input samples and associated input confidence values and has proprietary coefficients.

14. The filter of claim 13, wherein the digital filter comprises two digital filters, and wherein input samples are alternately allocated to one of the two digital filters.

15. A filter system, comprising:

the digital filter of any of the preceding claims 1-13, comprising a first digital filter and a second digital filter, each comprising an assigned filter function having assigned filter coefficients; a kind of electronic device with high-pressure air-conditioning system

A demultiplexer receives an input signal and generates input samples for the first digital filter and the second digital filter.

16. The filter system of claim 15, further comprising:

a first outlier detector that receives the input samples for the first digital filter and generates an associated input confidence value; a kind of electronic device with high-pressure air-conditioning system

A second outlier detector that receives the input samples for the second digital filter and generates an associated input confidence value.

17. The filter system of claim 15, wherein input samples for the first digital filter are samples around a first level and input samples for the second digital filter are samples around a second level, wherein the first level is higher than the second level.

18. A method of filtering digital input samples using a digital filter having an assigned filter function and comprising a set of filter coefficients, comprising the steps of:

receiving a digital input sample value and an associated input confidence value, wherein the input confidence value indicates a level of usefulness of the associated input sample;

Accumulating the input confidence values weighted with coefficients from the set of filter coefficients and generating a first accumulated value;

accumulating the input confidence values and generating a second accumulated value;

accumulating the input sample values and the input confidence values weighted with coefficients from the set of filter coefficients and generating a third accumulated value;

the confidence weighted input values are accumulated and a fourth accumulated value is generated.

19. The method as recited in claim 18, further comprising:

the first accumulated value is subtracted from a fixed value, wherein a result of the subtraction is divided by the second accumulated value and multiplied by the fourth accumulated value, and a result of the multiplication is added to the third accumulated value, and then the first accumulated value, the second accumulated value, the third accumulated value, and the fourth accumulated value are cleared.

20. The method of claim 19, wherein the fixed value is a sum of all coefficients.

21. The method of claim 18, wherein the input confidence value is binary.

22. A method of filtering digital input samples using a digital filter having an assigned filter function and comprising a first set of filter coefficients and a second set of filter coefficients defining a redistribution function, comprising the steps of:

accumulating the input confidence values weighted with coefficients from a first set of filter coefficients and generating a first accumulated value;

accumulating the input confidence values weighted with coefficients from a second set of filter coefficients and generating a second

An accumulated value;

accumulating input sample values weighted with coefficients from the first set of filter coefficients and the input confidence values, and generating a third accumulated value;

the input sample values and the input confidence values weighted with coefficients from the second set of filter coefficients are accumulated and a fourth accumulated value is generated.

23. The method of claim 22, further comprising subtracting the first accumulated value from a fixed value, wherein a result of subtracting is divided by the second accumulated value and multiplied by the fourth accumulated value, and a result of multiplying is added to the third accumulated value, and then clearing the first accumulated value, the second accumulated value, the third accumulated value, and the fourth accumulated value.

24. The method of claim 23, wherein the fixed value is a sum of all coefficients.

25. The method of claim 22, wherein the input confidence value is binary.