BLIND DISTORTION COMPENSATION FOR WIRELESS SYSTEMS

TECHNICAL FIELD The present disclosure relates mainly to wireless communication transceivers, such as radio base stations, microwave radio link transceivers, and other wireless devices. There are disclosed herein methods, circuits and signal processing devices for blind signal distortion mitigation, i.e., distortion suppressing techniques for mitigating linear and non-linear distortion which do not require a reference signal.

BACKGROUND

Radio receivers typically have stringent requirements on dynamic range, which in turn drive requirements on noise floor and residual distortion. Residual distortion typically comes from non-linear elements, like low noise amplifiers (LNAs), in the receiver chain. Other elements, like quadrature mixers and interleaved analog to digital converters (ADC), can also contribute with distortion in the form of signal images, which are unwanted spectral replicas.

Figure 1 illustrates an example 100 where a strong signal 110 has generated a plurality of distortion components, such as images 120 and non-linear distortion components 130. The signal 110 is comprised in a frequency band 115. A data signal 140, i.e., a signal carrying data to be received, is partly distorted by the distortion components.

Information of interest, such as user equipment (UE) traffic, radio base station (RBS) traffic and radio link signals, can reside both close to the noise floor (low input power) and far from the noise floor (high input power). One reason for this large dynamic range is that a receiver, such as an RBS, may cover a large physical area (e.g. a cell in a cellular access system), where wireless devices can be very close and very far away from the RBS at the same time. Unwanted radio signals with high input power are often denoted blockers or co-channel interferers. A blocker can be suppressed using known filtering techniques which suppresses signals in a given frequency band. However, with distortion present, there may be linear and non-linear images present as well. These images must also be suppressed, but since the images might coincide with information of interest, e.g., the data signal 140, filtering is not an option.

Figure 2 illustrates an example 200 where a data signal 210 is‘hidden’, i.e., severely distorted, by an image 220 generated by the strong blocker signal 1 10. Filtering cannot be applied to recover the data signal 210, since attenuating signals in the frequency band 215 would also attenuate the data signal 210.

Known distortion mitigation techniques comprise using a reference signal to separate out the wanted signal from the distortion. For instance, receive side distortion may be compensated using an in-band reference signal, such as a pre-amble known at the receiver. The portion of the received signal comprising the preamble can then be compared with the known preamble signal, and an error signal can be generated. This error signal can then be used to adapt an algorithm to compensate for the receive side distortion.

Transmit side distortion may be suppressed using a reference signal comprising the generated distortion. US20140192922 discloses using a separate transmit observation receiver to obtain the reference signal. However, obtaining the reference signal may not be so easy, and may drive both cost and communication overhead.

Known techniques also comprise so-called blind distortion mitigation techniques, which achieve distortion compensation without a reference signal. KR100703124 discloses one such example. However, most blind distortion compensation techniques are associated with reduced performance and/or slow convergence rates and may not be able to compensate for distortion with memory effect, i.e., when the present distortion depends not only on current signal values, but also on past signal values. There is a need for improved distortion mitigation techniques which do not rely on a-priori known reference signals or observation receivers.

SUMMARY It is an object of the present disclosure to provide improved devices, systems, and methods for distortion mitigation which do not require a known reference signal or observation receiver.

This object is at least in part obtained by a method for reducing distortion in a wireless system. The method comprises; obtaining a distorted signal s _dist comprising at least one dominating signal component, where each dominating signal component is comprised in a respective frequency band, and generating a distortion compensated signal ^s _comp by minimizing or reducing a difference between the distorted signal and an estimated reference signal s. The method also comprises generating the estimated reference signal s by filtering the compensated signal s _comp by a filter having a filter parameter that determines a frequency response of the filter, wherein the filter parameter is configured such that the filter amplifies signals in the at least one frequency band compared to signals outside of the at least one frequency band. This way distortion mitigation is performed without a reference signal, which is an advantage. Instead, a reference signal is estimated, whereupon non-blind distortion mitigation techniques can be re-used.

Memory effects can be compensated for in the proposed methods in a straight forward manner, which is an advantage not often associated with blind distortion mitigation methods. The method offers high performance and good convergence properties.

According to aspects, the at least one dominating signal component comprises a blocker signal, the respective frequency band of the blocker signal being located outside of a communication frequency band of the wireless system. Thus, advantageously, the disclosed methods are applicable in communication scenarios comprising blocker signals, such as in cellular access systems comprising radio base stations.

According to aspects, the at least one dominating signal component comprises a data signal, the respective frequency band of the data signal being comprised in a communication frequency band of the wireless system.

Thus, advantageously, the disclosed methods are also applicable in communication scenarios comprising data signal reception, such as in microwave radio links.

According to aspects, the generating comprises minimizing the difference between the distorted signal and the estimated reference signal in a least squares sense.

Least squares minimization offers a stable and robust optimization method associated with reasonable complexity, which is an advantage.

According to aspects, a filter parameter is determined such that the filter extracts the at least one dominating signal component from the distortion compensated signal s _comp .

Such filter parameters are obtainable in a straight forward manner without excessive complexity. It is an advantage that the reference signal estimator can be implemented as a linear filter, which is a low complexity implementation.

There are also disclosed herein computer programs, computer program products, signal processing devices, and wireless devices associated with the above-mentioned advantages.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure will now be described more in detail with reference to the appended drawings, where:

Figures 1 -2 show graphs illustrating signal components and distortion; Figure 3 shows a schematic view of a quadrature mixer;

Figure 4 schematically illustrates an analog to digital converter;

Figures 5-6 schematically illustrate distortion compensation systems;

Figures 7-8 schematically illustrate filtering operations;

Figures 9-10 schematically illustrate distortion compensation systems;

Figures 11 -13 show example wireless devices;

Figure 14 is a flow chart illustrating methods;

Figures 15-16 schematically illustrate signal processing devices; and

Figure 17 shows an example computer program product.

DETAILED DESCRIPTION

Aspects of the present disclosure will now be described more fully with reference to the accompanying drawings. The different devices and methods disclosed herein can, however, be realized in many different forms and should not be construed as being limited to the aspects set forth herein. Like numbers in the drawings refer to like elements throughout.

The terminology used herein is for describing aspects of the disclosure only and is not intended to limit the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

Herein, a frequency band denotes a continuous frequency range from a low frequency limit to a high frequency limit.

A power spectral density, or power density, is the power of a signal component per frequency unit. Power spectral density can be measured in, e.g., W/Hz. A signal component, unless otherwise noted, is assumed to be band-limited to a frequency band, which means that a main part of the signal energy of the signal component resides within the frequency band. The‘main part’ may be, e.g., a 99% bandwidth, a 3-dB bandwidth ora 6-dB bandwidth, i.e., a frequency limit where the signal spectral density has gone below some pre-defined threshold.

The disclosed techniques are mainly exemplified in discrete time, such as would be the case if the methods are applied to sampled signals. It is assumed that a sufficient sampling frequency is used to capture important signals, such as any dominating signal components. Analog versions and adaptations of the proposed techniques are straight forward.

In discrete time, signals may be represented as vectors, and processed by matrices. A vector comprises successive values of a discrete time signal in a known manner.

Herein, a vector may be denoted by lower case bold font, i.e., s _dist . The vector may be indexed by a subscript value n, i.e., s _{dist n} which is then a scalar value.

A matrix may herein be denoted by upper case bold font, i.e., H. A scalar value in a matrix is obtained by indexing using first and second index variables i 1 , and i2, i.e., H _{il i2} , where i1 denotes row index and i2 denotes column index.

It is appreciated that iterative techniques in general may be performed as batch processing or as sequential processing. Parameters may be estimated during system run-time or in advance during a start-up procedure similar to a calibration routine. There are introduced herein techniques for estimating parameters in a receiver distortion compensation model. The compensation model is used in a signal processing system to compensate for distortion in a received or transmitted signal. Reducing distortion may, in general, comprise suppressing linear and non-linear transforms of a wanted signal. Herein, reducing distortion may comprise, e.g., suppressing both linear and non-linear distortion, such as spectral growth, signal images, and the like.

Generally, the proposed techniques do not rely on any external reference, e.g. the methods do not rely on a non-distorted version of the signal of interest. Consequently, the proposed techniques belong to the class of blind distortion mitigation techniques. Receiver distortion may arise from linear and non-linear transformations in components and circuit elements , like (quadrature) mixers, amplifiers and (interleaved) ADCs. Linear distortion components, such as images, typically scale proportionally to the input signal. Consequently, images usually have constant Carrier to Interferer (C/I) power. Non-linear distortion typically scales non-linearly with the input signal. Hence, non-linear distortion C/I may improve as the blocker input power decreases.

Radio receivers often comprise hardware for down-converting a radio signal from a high frequency to a low frequency (an intermediate frequency) or to baseband. Figure 3 illustrates a complex base-band model 300 of a quadrature mixer which can be used for down-conversion. The model 300 shows a simplified receiver arranged for converting a Radio Frequency (RF) signal, s, centered around a frequency f _c down to another lower frequency f _c - f _L0 . The complex base-band model of the quadrature mixer shown in Figure 3 may comprise several imperfections, like the non-linear functions G, H,/H _Q and the additional phase shift f . The Local Oscillator (LO) mismatch modelled by the phase shift y, contributes to generation of 1 ^st order images (linear transforms), e.g., transforming a wanted signal s by A ^* conj(s), where A represents some amplitude scaling and conj() denotes complex conjugate. The non-linear distortions contribute with higher order images.

Analogue to Digital Converters (ADCs) with high sampling frequency typically comprise architectures with several time interleaved sub-ADCs. The sample rate of each sub-ADC is then lower than the ADC sample rate. The structure resembles a polyphase architecture (e.g. a polyphase filter). A time discrete base-band model 400 of an interleaved ADC is shown in Fell Hittar inte referenskalla.. Each phase of the polyphase structure, in the interleaved ADC model 400, may comprise two parts contributing to the distortion, namely direct current (DC) contribution and a non-linear filter contribution.

A model-based distortion compensation system according to prior art exploits parameterized models of elements contributing to the distortion, such as the mixer model 300 in Figure 3 and the ADC model 400 in Figure 4. The parameters of a model-based compensation must often be estimated since they are unknown. This estimation typically comprises minimizing some error or maximizing some probability, e.g., maximum likelihood (ML). In many cases, the square of an error is minimized, often denoted least squares (LS) minimization.

An example of a general compensation structure 500 is shown in Figure 5. Here, an input signal s is first subject to added noise , and then subject to a non-linear distortion function f(). More noise, n ₂ , is then added. A compensation function g() is aimed at countering the effects of the distortion. The compensated signal s _comp is then, if the compensation function g() resembles an inverse to function f(), close to the input signal s.

Using a model-based approach, a cost function or objective function may be formulated containing the unknown parameters of interest. An estimation algorithm then minimizes an error as defined by the cost function. The error could, e.g., be the difference between some wanted parameters and corresponding estimated parameters. The error can also be the difference between a compensated signal where distortion has been suppressed and a known reference signal.

For example, according to prior art, suppose a distorted signal is given by vector s _dist , and a reference signal vector s of some sort is available. For instance, the received signal may comprise a portion which is known at the receiver as discussed above. The receiver algorithm may then compare the known value with the received value and from there generate an error signal.

Suppose further that a compensation signal is given by Ha, where the observation matrix, H, contains a selection of time shifted linear and non-linear transformations of the distorted signal;

^dist [ > ^dist,-l> ^dist,Q> ^dist,l> ^dist,2> ] > H =

where ^* denotes complex conjugate, N is the length of the considered portion of vector s _dist , and where a is a parameter vector to be determined.

It is appreciated that matrix H can be populated by any number of linear or non-linear transforms of s _dist . The skilled person may determine a suitable matrix by experimentation or mathematical analysis or by computer simulation.

A cost function or objective function to be minimized over some range of a may then be given by J(a) = (s ₀ - s _dist - Ha) ^H (s ₀ - s _dist - Ha) , which is a LS-based objective function.

This cost function may also be weighted by a weighting matrix W, in which case a weighted least-squares objective function can be formed;

The solution to a (unweighted) LS minimization problem, i.e., an estimator for a, can be formulated as a = (H ^H H)- ¹ H ^H (s ₀ - s _dist ).

This estimator is a function of the distorted signal and the reference signal.

The compensated signal, where distortion has been suppressed, utilizing the estimated parameters, is then given by

As mentioned above, minimizing an error function typically means comparing some distortion compensated signal, like s _comp , with a known reference signal s ₀ or even a known part of the wanted signal s. In many systems, like transmitter Digital Pre-Distortion (DPD), there are transmitter observation receivers available (a receiver connected to the output of the power amplifier). A reference receiver could also be used when estimating parameters for receiver compensation. A reference receiver may for example sacrifice noise performance for improved non-linear performance, i.e., linearity. A reference receiver is designed to give a minimum of distortion which the compensation is intended to suppress, i.e. the reference receiver may have a different architecture compared to the main receiver. Such reference receivers are known and will not be discussed in more detail herein.

Introducing a reference receiver means extra analogue hardware (cost, power and area). Analogue hardware inherently carries some uncertainty due to the analogue components.

Analogue hardware may also be sensitive to unwanted internal and external leakage. Fell Hittar inte referenskalla. shows a system where an input signal s is first subject to noise n _common and a non-linearity as in Figure 5. The system in Figure 6 obtains a non-distorted version of the input signal s, to which a quantity of nose n _reference is added. This non-distorted version of s is then used in the compensation function to generate the compensated signal.

In the system 600 there might be leakage between the distorted signal and the reference receiver {s _dist might be present in s ₀ ), as illustrated by the dashed line 610. Such a leakage will limit the ability to perform unbiased estimation of compensation parameters.

The existence of reference receiver noise ( n _reference in Fell Hittar inte referenskalla.) typically means that extensive averaging is a necessity when estimating parameters, especially at low signal input powers. Extensive averaging may result in long convergence time and the estimator step response might be too long to follow changes of the parameters to be estimated, which is a drawback associated with the prior art. Note that, depending on the modeling, some parameters may depend on the input power of the signal. Thus, the hardware may be semi-stationary (slow changing), but a fast-varying input signal could be far from stationary. Such a situation requires the estimator to be fast in terms of convergence rate. Since the reference receivers introduce many problems, a new estimation method is of interest, which will be described in the following. Estimation algorithms not relying on a known reference are often referred to as blind estimators. Known blind estimators often minimize different cost functions compared to J(a ) above. Instead the cost function might be based on autocorrelation, power or mean values. Known blind estimators comprise estimators based, e.g., on constant modulus principles.

Blind estimators often utilize limitations of an existing system (or systemization). Such a limitation could for example be that the signal of interest is sparse in the frequency domain (e.g. bandwidth limited). If the signal of interest instead would be completely white, generally a blind estimator might be much more difficult to find.

The input signal, causing distortion, can often be assumed to be dominating;

Herein, dominating means having significantly higher signal power density in frequency domain compared to noise power density. It is appreciated that dominating does not necessarily mean to have significantly higher signal power compared to the power of the noise. Furthermore, the input signal causing distortion, could often be assumed bandwidth limited (sparse) compared to complete (digital) bandwidth. The distortion, caused by the dominating input signal, quite often falls outside the band of the distorted signal. Thus, the distortion to be suppressed may normally be observable in the frequency domain. The distortion could normally be assumed having much lower power spectral density compared to the dominating signal.

The distorted signal of interest, in discrete time indexed by time index n, could for example be described by

where the function, /, models the distortion parameterized by parameter vector a, and n _{l n} and n _{2 n} are (scalar) noise components, such as white Gaussian noise with some power spectral density. This model corresponds to the model 500 illustrated in Figure 5. An accurate reference signal for minimizing an error caused by the distortion would, e.g., be s _n . Had such a reference signal been available, the LS-based prior art methods discussed above could have been applied in a straight forward manner.

A blind algorithm does not have access to a reference signal. Instead, an estimate of the values of s _n for some range of n, herein denoted s _n is used. It has been realized that one method to find s _n is to utilize

s = H _est( p, where f is a parameter vector, and the‘observation matrix’ H _est comprises time shifted and vectorized versions of s _{comp n} \

It is appreciated that matrix H _est can be populated by any number of linear (or non-linear) transforms of vector s _comp . The skilled person may determine a suitable matrix by, e.g., experimentation or mathematical analysis or by computer simulation. The size of matrix H _est , i.e., the number of rows and columns, can be varied depending on application and scenario.

The size of H _est contributes to the overall computational complexity of the proposed methods. It is therefore preferred to keep the size of H _est as small as possible.

The proposed method is advantageously implemented as an iterative method, where the parameter vector f is determined iteratively. Herein, such iterations are indexed by a variable m. Thus, denotes the value of f at iteration m.

The compensated signal, s _comp , is defined by ^s _co m _p = Ha + s _dist , which is a vector having the same length as s _dist , and where the observation matrix, H, again contains time shifted linear and non- linear transformations of the distorted signal s _dist , as exemplified above, and where a is a parameter vector also like in the above discussions. The refence estimator (H _est <p), which represents the blind estimate of a reference signal, is herein proposed to be a filter such as a linear transform of the distortion compensated signal s _dist .

An example method for finding f is to define a signal time index, and h _h is Gaussian noise with zero mean and variance s* . s _{comp n} is the distortion mitigated signal at time index n.

A threshold in the frequency domain can be used to generate a band pass filter to be used in the time domain (i.e. estimating ^f in the frequency domain by setting the gain to 1 for spectral power above the threshold and 0 or some small number below). The filtering can also be performed directly in frequency domain using known techniques. An example of the signals r _n and s _{comp n} is shown in Fell Hittar inte referenskalla. assuming first iteration and therefore no compensation. Note that the compensated signal s _{comp n} may be a blocker signal, or a data signal comprising information to be received.

The reference estimator parameter vector f can be found by minimizing H _es t, noise <p) ( s _comp - H _{est Jl0ise (} p), over some suitable range of f, where H est, noise contains time shifted version of r =

It is appreciated that matrix H _{est noise} can be populated by any number of linear (or non-linear) transforms of r. The skilled person may determine a suitable matrix by experimentation or mathematical analysis or by computer simulation.

An example of an estimated reference signal, s (assuming the role of s or s ₀ above), based on Fell Hittar inte referenskalla. is schematically illustrated in Fell Hittar inte referenskalla..

Finding the reference estimator filter and the compensation parameters and respectively) is advantageously performed in an iterative process (thereof the index m, as introduced above). The main idea behind the proposed method is to find a reference which is at least slightly better (closer to s _n ) compared to s _{dist 7l} . Solving

would then result in a compensated signal at the second iteration (where m=2);

slightly better than the distorted signal s _dist . In the first iteration, where m = 1, the estimated compensation values can be set to = 0, i.e., the all-zero vector. The estimated compensation values can also be set to some other suitable value. The proposed blind reference estimator structure 900 is illustrated in Fell Hittar inte referenskalla..

In Figure 9, an input signal s _n (with time index n) is first subject to noise 910 before being subjected to a distortion transform f() 920 parameterized by parameter a, which may well comprise non-linear distortion. The distorted signal is then again subject to noise 930. The noise may be additive white Gaussian noise (AWGN), or some other type of noise.

The distorted signal enters a compensation function 940, which generates the compensated signal s _{comp n} .

The compensated signal is fed back to the reference estimation function, where noise is added 950. This noisy compensated signal is used as input for the reference estimating function 960, 970 which generates the filter parameters f^ , and the estimated reference signal s. The different matrix functions have been indicated in the respective function blocks.

The proposed method has so-far been presented as a technique to suppress distortion caused by non-linear elements in radio receivers. The method may also be used in any scenario resembling the problems related to non-linear distortion in radio receivers, such as any type of unknown signal (radio, electrical, optical, etc.) exposed to distortion. There is at least one further scenario where the proposed method is applicable. This scenario contains an adaptive transmitter pre-distortion with the intention to compensate a transmitted signal exposed to non-linear distortion, where the estimator of the pre-distorter does not require any reference signal. A typical pre-distorter used in such a configuration would be a Digital Pre-Distorter (DPD). One important special case is when an Analogue Pre-Distorter (APD) is used instead. An APD does not operate on the signal in the digital domain but may still have configurable parameters. Such a scenario is shown in Figure 10.

In Figure 10, an APD processes an input signal s _in in order to pre-distort it. This pre-distortion compensates in advance for any transforms the signal will be subject to before exiting the system as a transmit signal s _tx . In Figure 10, the pre-distorted signal s _DPD is first up-converted in frequency by means of a mixer 1010, and filtered by a filter 1020, here a band-pass filter. The signal is then amplified by a power amplifier (PA). Both the mixer, the filter, and the PA may transform the input signal. Flowever, if the pre-distorted signal s _DPD is correctly generated, the transmitted signal s _tx will still resemble the input signal.

To estimate the DPD parameters, the transmitted signal is fed back and down- converted in frequency by another mixer 1030. A second filter 1040 is applied, here a low-pass filter. The filtered signal 1045 is then used as reference signal s ₀ to the blind DPD parameter estimation.

An APD could for example be implemented as Generalized Memory Polynomial (GMP). The behavior of a GMP-APD would be equivalent to corresponding GMP-DPD, with some unknown parameters to be estimated (e.g. polynomial coefficients). Using a transmitter observation receiver, the proposed method could be used to estimate the unknown parameters of said APD. The proposed method could also be used for estimating parameters in a DPD with the advantage that the estimator does not require to be phase and/or frequency locked to the transmit signal ( s _in ). A distortion estimator, which does not require complete knowledge of an undistorted version of the signal (reference), may simplify the transmitter radio architecture. Thus, the main potential advantages of this proposed solution are cost, money and power.

Figures 1 1 -13 show further example wireless systems and devices where the proposed distortion mitigation techniques may be applied. Figure 1 1 shows a point to point radio link 1 100 between a first transceiver 1 1 10 and a second transceiver 1 120. The proposed techniques may be used to mitigate both transmit side and receive side distortion in the first and second transceivers. Figure 12 shows a cellular access system 1200 where an RBS 1210 is arranged to serve a plurality of wireless device 1220. The proposed techniques may be used to mitigate transmit and receive side distortion in both the RBS and in the wireless devices. Figure 13 shows a general transceiver where the proposed techniques are used to mitigate transmit distortion by an APD such as that schematically shown in Figure 10.

Figure 14 is a flow chart illustrating methods. The methods summarize the discussions above.

There is illustrated a method for reducing distortion in a wireless system. The method is preferably but not necessarily an iterative method where a time index n denotes discrete time and an iteration index m denotes iteration number. It is appreciated that continuous time extensions of the method are straight forward. The method comprises obtaining S1 a distorted signal comprising at least one dominating signal component 1 10. The dominating signal component may be a strong blocker signal or a data signal, as illustrated in Figures 1 and 2, or a combination of blocker and data signals. The distorted signal was referred to in the examples above as s _dist .

Each dominating signal component is comprised in a respective frequency band 1 15. It is assumed that the respective frequency band is available to the method, i.e., that the digital bandwidth is sufficiently large to comprise the dominating signal component.

The method also comprises generating S2 a distortion compensated signal ^s _co m _p by minimizing a difference between the distorted signal and an estimated reference signal s. The distortion compensated signal was referred to above as vector s _comp , and the estimated reference signal was referred to above as vector s.

It is appreciated that the‘difference’ which is minimized can be defined in a number of different ways, resulting in slightly different minimization criteria. For instance, the difference can be minimized in a least-squares sense. The difference can also be minimized according to other criteria, such as maximum likelihood (ML), or maximum a-posteriori (MAP).

Herein,‘to minimize’ may also comprise‘to reduce’. Thus, it is appreciated that the difference need not necessarily be minimized to a minimum value. It is, according to some aspects, sufficient to significantly reduce the difference value down to some acceptable level.

The method also comprises generating S3 the estimated reference signal s by filtering the compensated signal s _comp by a filter having a filter parameter that determines a frequency response of the filter. The filter parameter was referred to above as f. The filter frequency response was exemplified in, e.g., Figures 7 and 8.

Notably, the filter parameter is configured such that the filter amplifies signals in the at least one frequency band 115 compared to signals outside of the at least one frequency band 115.

Advantageously, the method does not require a reference signal. Instead, the reference signal is estimated by the method.

The method can be executed in a batch mode, where a portion of obtained distorted signal is processed in order to generate the distortion compensated signal. The method then uses the same time section of distorted signal and gradually converges towards a distortion compensated signal where distortion has been suppressed.

The method may also be executed in a time sequential mode, where new samples of the distorted signal is obtained successively. The method then processes the new samples as they enter the system, while gradually converging in the distortion compensation parameters to generate the distortion compensated signal.

Batch mode and sequential mode can be combined by sequentially processing new batches of data.

It is appreciated that the method comprises variables which may need to be initialized at some initialization value. The skilled person realizes that such initialization may be performed in different ways. For instance, the initialization may be performed at some fixed value, such = 0 . The initialization may also be performed using a value determined in simulations, or by computer experimentation, or using a previous value obtained in previous runs of the method with a similar system.

Consequently, the method may comprise a step of initialization of system parameters and variables prior to generating the first distortion compensated signal, as shown in Figure 14.

According to some aspects, the at least one dominating signal component comprises a blocker signal, the respective frequency band of the blocker signal being located outside of a communication frequency band of the wireless system.

According to some other aspects, the at least one dominating signal component comprises a data signal, the respective frequency band of the data signal being comprised in a communication frequency band of the wireless system.

The frequency band where the dominating signal component is comprised may be detected using known frequency domain methods, such as thresholding and by applying weights. For instance, in case the dominating signal is a data signal to be received, then the frequency band can be assumed known. In case the dominating signal component is a strong blocker, then frequency domain techniques can be used to identify the blocker, and to determine the frequency band where the strong blocker resides. Filtering can also be performed in frequency domain directly, i.e., filtering the compensated signal s _comp by the filter having a filter parameter that determines a frequency response of the filter can be performed using frequency domain filtering techniques.

The frequency band may also be assumed known a-priori or taken as input to the method.

The frequency band may also be implicitly detected, e.g., by adding noise as in Figure 7. In this case the location of the frequency band need not be known, only the required power spectral density of the noise.

According to aspects, the generating S21 comprises minimizing the difference between the distorted signal and the estimated reference signal in a least squares sense. This minimization may comprise applying a least-mean- squares (LMS) algorithm, or a recursive least squares (RLS) algorithm. The difference can be minimized using batch processing of a sampled signal or minimized during system run-time. According to the above description, the method may be initialized according to a pre-determ ined initialization parameter.

According to aspects, the filter parameter is determined such that the filter extracts the at least one dominating signal component 110 from the distortion compensated signal s _comp .

This means that the dominating signal component or components are emphasized compared to the noise and the distortion, and also compared to any other signal components present. It is appreciated that to extract may mean to amplify the dominating signal component, or to suppress other signal components.

According to aspects, as discussed above, the filter parameter is a vector parameter f determined by minimizing objective function /(<p),

where H _{est noise} contains time shifted versions of ^r n = s _{comp n} + h _h h _h ~ N{0, s*), and is set below a power spectral density associated with the at least one dominating component. H _est, noise may also comprise time-shifted non-linear transforms of the signal r _n -

According to aspects, as discussed above, the filter parameter is determined based on a configured threshold in the frequency domain, wherein the threshold is set at a level below the power spectral density of the at least one dominating component and above the power spectral density of at least some distortion components, wherein signal content below the threshold is scaled down or set to zero.

According to aspects, as discussed above, the filter parameter is determined to generate a band pass filter for use in time domain or in frequency domain, wherein the band pass filter is arranged to extract the at least one dominating component.

According to aspects, the distorted signal s _dist comprises distortion from a quadrature mixer, such as the mixer 300 schematically illustrated in Figure 3.

According to aspects, the distorted signal s _dist comprises distortion from an interleaved analog-to-digital converter, ADC, such as the ADC 400 schematically illustrated in Figure 4.

According to aspects, the distorted signal s _dist comprises distortion from a power amplifier, such as the power amplifier schematically illustrated in Figure 10, or a low-noise amplifier (LNA).

According to aspects, as discussed above, a matrix H contains time shifted linear and non-linear transformations of the distorted signal s _dist , s is the estimated reference signal, a - s _dist ), and the distortion compensated signal is given by

According to aspects, as discussed above, in a first iteration, = 0.

Figure 15 schematically illustrates, in terms of a number of functional units, the components of a distortion mitigation system 900 according to an embodiment of the discussions herein. Processing circuitry 1510 is provided using any combination of one or more of a suitable central processing unit CPU, multiprocessor, microcontroller, digital signal processor DSP, etc., capable of executing software instructions stored in a computer program product, e.g. in the form of a storage medium 1530. The processing circuitry 1510 may further be provided as at least one application specific integrated circuit ASIC, or field programmable gate array FPGA. The processing circuitry thus comprises a plurality of digital logic components.

Particularly, the processing circuitry 1510 is configured to cause the distortion mitigation system 900 to perform a set of operations, or steps. For example, the storage medium 1530 may store the set of operations, and the processing circuitry 1510 may be configured to retrieve the set of operations from the storage medium 1530 to cause the distortion mitigation system 900 to perform the set of operations. The set of operations may be provided as a set of executable instructions. Thus, the processing circuitry 1510 is thereby arranged to execute methods as herein disclosed.

The storage medium 1530 may also comprise persistent storage, which, for example, can be any single one or combination of magnetic memory, optical memory, solid state memory or even remotely mounted memory.

The distortion mitigation system 900 further comprises an interface 1520 for communications with at least one external device, such as the ADC 400 discussed in connection to Figure 4. As such the interface 1520 may comprise one or more transmitters and receivers, comprising analogue and digital components and a suitable number of ports for wireline or wireless communication.

The processing circuitry 1510 controls the general operation of any comprised transceiver, e.g. by sending data and control signals to the interface 1520 and the storage medium 1530, by receiving data and reports from the interface 1520, and by retrieving data and instructions from the storage medium 1530. Other components, as well as the related functionality, of the control node are omitted in order not to obscure the concepts presented herein. According to the discussion above, Figure 15 schematically illustrates a signal processing device 900 arranged to reduce distortion in a wireless system. The signal processing device comprises

an obtaining module Sx1 configured to obtain a distorted signal s _dist comprising at least one dominating signal component 110, where each dominating signal component is comprised in a respective frequency band 115,

a first generating module Sx2 configured to generate a distortion compensated signal s _comp by minimizing a difference between the distorted signal and an estimated reference signal s, and

a second generating module Sx3 configured to generate the estimated reference signal s by filtering the compensated signal s _comp by a filter having a filter parameter that determines a frequency response of the filter, wherein the filter parameter is configured such that the filter amplifies signals in the at least one frequency band compared to signals outside of the at least one frequency band.

According to aspects, the first generating module is configured to minimize the difference between the distorted signal and the estimated reference signal in a least squares sense.

According to aspects, the filter parameter is a vector parameter f determined by minimizing objective function /(<p),

where H _{est noise} contains time shifted versions of ^r n s _{comp n} + h _h h _h ~ N{0, s*), and is set below a power spectral density associated with the at least one dominating component.

H _{est, noise} may also comprise time-shifted non-linear transforms of the signal According to aspects, a matrix H contains time shifted linear and non-linear transformations of the distorted signal s _dist , s is the estimated reference signal, a = - s _dist ), and the distortion compensated signal is given by

According to aspects, as discussed above, in a first iteration, = 0.

Figure 16 schematically shows the different modules Sx1 , Sx2, Sx21 , Sx3.

Figure 17 schematically illustrates a computer program product 1700 comprising computer program code 1710 which, when executed in a distortion mitigation system such as that discussed herein causes the system to execute a distortion mitigation method.